feat(dm): добавлены учебные задания витрин и даг самопроверки

- Зачем:
  - менти строит витрины DM сам, от DDL по правилу зерна до задачи в
    конвейере; стенду нужны ТЗ с образцом и обратная связь по каждой
    витрине (#159, резолюция #135).
- Что:
  - docs/labs/: README и пять ТЗ — events_enriched_v, top_pages_daily,
    session_overview, dq_errors_daily, utm_effectiveness; каждое ТЗ
    называет урок, состав колонок, форму по зерну и самопроверку.
  - dags/homework_check.py: ручной даг, задача на витрину, красная без
    витрины или при расхождении; проверки — шесть файлов sql/dq/ с
    инвариантами против DDS и ODS, без золотых констант; в dq_summary
    не пишет.
  - docs: dm.md называет homework_check и снимает «Остается нарезке»,
    §7 и §10 мастер-спеки согласованы с лабами, dds/session.md указывает
    на ТЗ, AGENTS.md пополнен правилом имен docs/labs/.
- Проверка:
  - make lint; make smoke 21/21, make check-clickhouse 10/10,
    make check-services 7/7.
  - на свежем стенде все пять задач homework_check красные и называют
    ТЗ; черновик каждой витрины по тексту ТЗ зеленит свою задачу,
    после сноса — снова красная; строк от самопроверки в dq_summary нет.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

sql/dq/dq_errors_daily_vs_ods.sql

@@ -0,0 +1,88 @@
+-- DQ: дневная сводка брака против живого окна таблиц ошибок ODS.
+--
+-- Граница окна повторяет TTL источников; более старую историю сверять не с чем.
+
+WITH
+    reference AS
+    (
+        SELECT
+            load_date,
+            source,
+            error_class,
+            count() AS error_rows
+        FROM
+        (
+            SELECT
+                toDate(_load_ts) AS load_date,
+                'event' AS source,
+                error_class
+            FROM ods.event_errors_dist
+            UNION ALL
+            SELECT
+                toDate(_load_ts) AS load_date,
+                'order' AS source,
+                error_class
+            FROM ods.order_errors_dist
+        )
+        WHERE load_date >= toDate(now() - INTERVAL 1 MONTH)
+        GROUP BY load_date, source, error_class
+    ),
+    actual AS
+    (
+        SELECT
+            load_date,
+            source,
+            error_class,
+            count() AS physical_rows,
+            any(error_rows) AS error_rows
+        FROM dm.dq_errors_daily_v
+        WHERE load_date >= toDate(now() - INTERVAL 1 MONTH)
+        GROUP BY load_date, source, error_class
+    ),
+    comparison AS
+    (
+        SELECT
+            coalesce(reference.load_date, actual.load_date) AS data_date,
+            concat(
+                coalesce(reference.source, actual.source),
+                '/',
+                coalesce(reference.error_class, actual.error_class)
+            ) AS business_key,
+            isNotNull(reference.source) AS reference_present,
+            isNotNull(actual.source) AS actual_present,
+            reference.error_rows AS reference_value,
+            actual.error_rows AS actual_value,
+            actual.physical_rows AS physical_rows,
+            isNull(reference.source)
+                OR isNull(actual.source)
+                OR actual.physical_rows != 1
+                OR NOT isNotDistinctFrom(reference.error_rows, actual.error_rows)
+                AS failed
+        FROM reference
+        GLOBAL FULL OUTER JOIN actual USING (load_date, source, error_class)
+    )
+SELECT
+    data_date,
+    countIf(reference_present) AS reference_rows,
+    countIf(actual_present) AS actual_rows,
+    countIf(failed) AS failed_rows,
+    groupArrayIf(20)(
+        concat(
+            toString(data_date),
+            ' source/error_class=',
+            business_key,
+            ': источник=',
+            if(reference_present, toString(reference_value), 'нет ключа'),
+            ', витрина=',
+            if(
+                actual_present,
+                concat(toString(actual_value), ', rows=', toString(physical_rows)),
+                'нет ключа'
+            )
+        ),
+        failed
+    ) AS diagnostics
+FROM comparison
+GROUP BY data_date
+ORDER BY data_date
+SETTINGS join_use_nulls = 1

sql/dq/events_enriched_scope.sql

@@ -0,0 +1,15 @@
+-- DQ: область посуточной сверки обогащенных событий.
+--
+-- Объединение сторон оставляет видимым день, потерянный целиком.
+
+SELECT day
+FROM
+(
+    SELECT event_date AS day
+    FROM dds.event_v
+    UNION ALL
+    SELECT event_date AS day
+    FROM dm.events_enriched_v
+)
+GROUP BY day
+ORDER BY day

sql/dq/events_enriched_vs_dds.sql

@@ -0,0 +1,125 @@
+-- DQ: обогащенные события против прямого ответа DDS за один день.
+--
+-- День передает даг, чтобы широкая сверка не собирала всю историю в памяти.
+
+WITH
+    reference AS
+    (
+        SELECT
+            event_date,
+            event_id,
+            tuple(
+                event_time,
+                toHour(event_time),
+                client_id,
+                event_type,
+                url,
+                path(url),
+                referer,
+                domain(referer),
+                title,
+                utm_source,
+                utm_medium,
+                utm_campaign,
+                utm_content,
+                utm_term,
+                device_category,
+                browser,
+                operating_system,
+                region_country,
+                region_city,
+                product_id,
+                product_name,
+                product_category,
+                product_price,
+                product_quantity,
+                product_event_type,
+                purchase_id,
+                purchase_revenue,
+                purchase_coupon
+            ) AS value
+        FROM dds.event_v
+        WHERE event_date = {day:Date}
+    ),
+    actual AS
+    (
+        SELECT
+            event_date,
+            event_id,
+            count() AS physical_rows,
+            any(tuple(
+                event_time,
+                event_hour,
+                client_id,
+                event_type,
+                url,
+                page_path,
+                referer,
+                referer_host,
+                title,
+                utm_source,
+                utm_medium,
+                utm_campaign,
+                utm_content,
+                utm_term,
+                device_category,
+                browser,
+                operating_system,
+                region_country,
+                region_city,
+                product_id,
+                product_name,
+                product_category,
+                product_price,
+                product_quantity,
+                product_event_type,
+                purchase_id,
+                purchase_revenue,
+                purchase_coupon
+            )) AS value
+        FROM dm.events_enriched_v
+        WHERE event_date = {day:Date}
+        GROUP BY event_date, event_id
+    ),
+    comparison AS
+    (
+        SELECT
+            coalesce(reference.event_date, actual.event_date) AS data_date,
+            coalesce(reference.event_id, actual.event_id) AS business_key,
+            isNotNull(reference.event_id) AS reference_present,
+            isNotNull(actual.event_id) AS actual_present,
+            reference.value AS reference_value,
+            actual.value AS actual_value,
+            actual.physical_rows AS physical_rows,
+            isNull(reference.event_id)
+                OR isNull(actual.event_id)
+                OR actual.physical_rows != 1
+                OR NOT isNotDistinctFrom(reference.value, actual.value) AS failed
+        FROM reference
+        GLOBAL FULL OUTER JOIN actual USING (event_date, event_id)
+    )
+SELECT
+    data_date,
+    countIf(reference_present) AS reference_rows,
+    countIf(actual_present) AS actual_rows,
+    countIf(failed) AS failed_rows,
+    groupArrayIf(20)(
+        concat(
+            toString(data_date),
+            ' event_id=',
+            toString(business_key),
+            ': источник=',
+            if(reference_present, toString(reference_value), 'нет ключа'),
+            ', витрина=',
+            if(
+                actual_present,
+                concat(toString(actual_value), ', rows=', toString(physical_rows)),
+                'нет ключа'
+            )
+        ),
+        failed
+    ) AS diagnostics
+FROM comparison
+GROUP BY data_date
+ORDER BY data_date
+SETTINGS join_use_nulls = 1

sql/dq/session_overview_vs_dds.sql

@@ -0,0 +1,87 @@
+-- DQ: дневной портрет сессий против прямого агрегата сессий DDS.
+--
+-- Полное объединение не прячет день или устройство, исчезнувшие с любой стороны.
+
+WITH
+    reference AS
+    (
+        SELECT
+            session_date AS report_date,
+            device_category,
+            tuple(
+                count(),
+                uniqExact(client_id),
+                countIf(events_total = 1),
+                countIf(purchases > 0),
+                sum(duration_seconds),
+                sum(events_total),
+                sum(pageviews),
+                sum(cart_adds),
+                sum(purchases)
+            ) AS value
+        FROM dds.session_v
+        GROUP BY report_date, device_category
+    ),
+    actual AS
+    (
+        SELECT
+            report_date,
+            device_category,
+            count() AS physical_rows,
+            any(tuple(
+                sessions,
+                visitors,
+                bounce_sessions,
+                purchase_sessions,
+                duration_seconds_total,
+                events_total,
+                pageviews,
+                cart_adds,
+                purchases
+            )) AS value
+        FROM dm.session_overview_v
+        GROUP BY report_date, device_category
+    ),
+    comparison AS
+    (
+        SELECT
+            coalesce(reference.report_date, actual.report_date) AS data_date,
+            coalesce(reference.device_category, actual.device_category)
+                AS business_key,
+            isNotNull(reference.device_category) AS reference_present,
+            isNotNull(actual.device_category) AS actual_present,
+            reference.value AS reference_value,
+            actual.value AS actual_value,
+            actual.physical_rows AS physical_rows,
+            isNull(reference.device_category)
+                OR isNull(actual.device_category)
+                OR actual.physical_rows != 1
+                OR NOT isNotDistinctFrom(reference.value, actual.value) AS failed
+        FROM reference
+        GLOBAL FULL OUTER JOIN actual USING (report_date, device_category)
+    )
+SELECT
+    data_date,
+    countIf(reference_present) AS reference_rows,
+    countIf(actual_present) AS actual_rows,
+    countIf(failed) AS failed_rows,
+    groupArrayIf(20)(
+        concat(
+            toString(data_date),
+            ' device_category=',
+            business_key,
+            ': источник=',
+            if(reference_present, toString(reference_value), 'нет ключа'),
+            ', витрина=',
+            if(
+                actual_present,
+                concat(toString(actual_value), ', rows=', toString(physical_rows)),
+                'нет ключа'
+            )
+        ),
+        failed
+    ) AS diagnostics
+FROM comparison
+GROUP BY data_date
+ORDER BY data_date
+SETTINGS join_use_nulls = 1

sql/dq/top_pages_daily_vs_dds.sql

@@ -0,0 +1,71 @@
+-- DQ: дневная популярность страниц против прямого агрегата событий DDS.
+--
+-- Полное объединение не прячет страницу, исчезнувшую с любой стороны.
+
+WITH
+    reference AS
+    (
+        SELECT
+            event_date AS report_date,
+            path(url) AS page_path,
+            tuple(
+                countIf(event_type = 'pageview'),
+                uniqExact(client_id),
+                countIf(event_type = 'add_to_cart'),
+                countIf(event_type = 'purchase')
+            ) AS value
+        FROM dds.event_v
+        GROUP BY report_date, page_path
+    ),
+    actual AS
+    (
+        SELECT
+            report_date,
+            page_path,
+            count() AS physical_rows,
+            any(tuple(pageviews, visitors, cart_adds, purchases)) AS value
+        FROM dm.top_pages_daily_v
+        GROUP BY report_date, page_path
+    ),
+    comparison AS
+    (
+        SELECT
+            coalesce(reference.report_date, actual.report_date) AS data_date,
+            coalesce(reference.page_path, actual.page_path) AS business_key,
+            isNotNull(reference.page_path) AS reference_present,
+            isNotNull(actual.page_path) AS actual_present,
+            reference.value AS reference_value,
+            actual.value AS actual_value,
+            actual.physical_rows AS physical_rows,
+            isNull(reference.page_path)
+                OR isNull(actual.page_path)
+                OR actual.physical_rows != 1
+                OR NOT isNotDistinctFrom(reference.value, actual.value) AS failed
+        FROM reference
+        GLOBAL FULL OUTER JOIN actual USING (report_date, page_path)
+    )
+SELECT
+    data_date,
+    countIf(reference_present) AS reference_rows,
+    countIf(actual_present) AS actual_rows,
+    countIf(failed) AS failed_rows,
+    groupArrayIf(20)(
+        concat(
+            toString(data_date),
+            ' page_path=',
+            business_key,
+            ': источник=',
+            if(reference_present, toString(reference_value), 'нет ключа'),
+            ', витрина=',
+            if(
+                actual_present,
+                concat(toString(actual_value), ', rows=', toString(physical_rows)),
+                'нет ключа'
+            )
+        ),
+        failed
+    ) AS diagnostics
+FROM comparison
+GROUP BY data_date
+ORDER BY data_date
+SETTINGS join_use_nulls = 1

sql/dq/utm_effectiveness_vs_dds.sql

@@ -0,0 +1,88 @@
+-- DQ: эффективность UTM-меток против прямого агрегата событий DDS.
+--
+-- Полное объединение не прячет набор меток, исчезнувший с любой стороны.
+
+WITH
+    reference AS
+    (
+        SELECT
+            event_date AS report_date,
+            utm_source,
+            utm_medium,
+            utm_campaign,
+            tuple(
+                uniqExact(client_id),
+                countIf(event_type = 'pageview'),
+                countIf(event_type = 'add_to_cart'),
+                countIf(event_type = 'purchase'),
+                sumIf(purchase_revenue[1], event_type = 'purchase')
+            ) AS value
+        FROM dds.event_v
+        GROUP BY report_date, utm_source, utm_medium, utm_campaign
+    ),
+    actual AS
+    (
+        SELECT
+            report_date,
+            utm_source,
+            utm_medium,
+            utm_campaign,
+            count() AS physical_rows,
+            any(tuple(
+                visitors,
+                pageviews,
+                add_to_carts,
+                purchases,
+                declared_revenue
+            )) AS value
+        FROM dm.utm_effectiveness_v
+        GROUP BY report_date, utm_source, utm_medium, utm_campaign
+    ),
+    comparison AS
+    (
+        SELECT
+            coalesce(reference.report_date, actual.report_date) AS data_date,
+            concat(
+                'utm_source=', coalesce(reference.utm_source, actual.utm_source),
+                ', utm_medium=', coalesce(reference.utm_medium, actual.utm_medium),
+                ', utm_campaign=',
+                coalesce(reference.utm_campaign, actual.utm_campaign)
+            ) AS business_key,
+            isNotNull(reference.utm_source) AS reference_present,
+            isNotNull(actual.utm_source) AS actual_present,
+            reference.value AS reference_value,
+            actual.value AS actual_value,
+            actual.physical_rows AS physical_rows,
+            isNull(reference.utm_source)
+                OR isNull(actual.utm_source)
+                OR actual.physical_rows != 1
+                OR NOT isNotDistinctFrom(reference.value, actual.value) AS failed
+        FROM reference
+        GLOBAL FULL OUTER JOIN actual
+            USING (report_date, utm_source, utm_medium, utm_campaign)
+    )
+SELECT
+    data_date,
+    countIf(reference_present) AS reference_rows,
+    countIf(actual_present) AS actual_rows,
+    countIf(failed) AS failed_rows,
+    groupArrayIf(20)(
+        concat(
+            toString(data_date),
+            ' ',
+            business_key,
+            ': источник=',
+            if(reference_present, toString(reference_value), 'нет ключа'),
+            ', витрина=',
+            if(
+                actual_present,
+                concat(toString(actual_value), ', rows=', toString(physical_rows)),
+                'нет ключа'
+            )
+        ),
+        failed
+    ) AS diagnostics
+FROM comparison
+GROUP BY data_date
+ORDER BY data_date
+SETTINGS join_use_nulls = 1