feat(openvino): реализован OpenVINO бэкенд транскрипции

- Зачем:
  - ускорение транскрипции на x86 CPU (Intel/AMD) в 2-4 раза через OpenVINO GenAI.
- Что:
  - создан backends/openvino.py: OpenVINOBackend с ensure_model_available, create_model, transcribe.
  - модели скачиваются из HuggingFace (OpenVINO/whisper-*-ov), формат OpenVINO IR.
  - аудио декодируется через faster_whisper.decode_audio (PyAV) → .tolist() → pipe.generate(return_timestamps=True).
  - контракт compute_type: явный --compute-type уважается; из дефолтов large-v3 получает fp16 автоматически.
  - openvino-genai добавлен в pyproject.toml с platform markers (x86_64/AMD64, не macOS).
  - compute_type_explicit прокинут через get_backend → load_model → CLI.
  - 16 тестов для OpenVINO бэкенда: resolve_repo, ensure, create, transcribe, validate.
- Проверка:
  - uv run pytest -v — 119 passed.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

tests/test_transcriber.py

@@ -124,7 +124,7 @@ def test_transcribe_cuda_fallback(mock_get_backend):
         model_path="/mock/cpu/model",
     )
 
-    def backend_for_device(device):
+    def backend_for_device(device, **kwargs):
         return cuda_backend if device == "cuda" else cpu_backend
 
     mock_get_backend.side_effect = backend_for_device
@@ -168,7 +168,7 @@ def test_transcribe_cuda_fallback_on_transcribe_call(mock_get_backend):
         model_path="/mock/cpu/model",
     )
 
-    def backend_for_device(device):
+    def backend_for_device(device, **kwargs):
         return cuda_backend if device == "cuda" else cpu_backend
 
     mock_get_backend.side_effect = backend_for_device
@@ -245,7 +245,7 @@ def test_transcribe_non_strict_cuda_fallback(mock_get_backend):
         model_path="/mock/cpu/model",
     )
 
-    def backend_for_device(device):
+    def backend_for_device(device, **kwargs):
         return cuda_backend if device == "cuda" else cpu_backend
 
     mock_get_backend.side_effect = backend_for_device
@@ -288,7 +288,7 @@ def test_load_model_cuda_fallback(mock_get_backend):
     cpu_model = MagicMock()
     cpu_backend = _make_backend(model=cpu_model, model_path="/mock/cpu/model")
 
-    def backend_for_device(device):
+    def backend_for_device(device, **kwargs):
         return cuda_backend if device == "cuda" else cpu_backend
 
     mock_get_backend.side_effect = backend_for_device