feat(US3+4): video frame extraction + video-to-text — POST /api/v1/video/*

- app/models/video_models.py: ExtractFramesRequest, VideoToTextRequest,
  FrameInfo, VideoJobCallback, VideoAcceptedResponse
- app/services/video_service.py: interval+keyframe frame extraction,
  uniform-sample video-to-text, HTTP callback, temp file cleanup
- app/routers/video.py: size check helper (_check_video_size via head_object),
  BackgroundTasks enqueue for both endpoints
- tests: 6 service + 4 router tests, 10/10 passing

app/models/video_models.py

@@ -0,0 +1,38 @@
+from pydantic import BaseModel
+
+
+class ExtractFramesRequest(BaseModel):
+    file_path: str
+    source_id: int
+    job_id: int
+    mode: str = "interval"
+    frame_interval: int = 30
+
+
+class VideoToTextRequest(BaseModel):
+    file_path: str
+    source_id: int
+    job_id: int
+    start_sec: float
+    end_sec: float
+    model: str | None = None
+    prompt_template: str | None = None
+
+
+class FrameInfo(BaseModel):
+    frame_index: int
+    time_sec: float
+    frame_path: str
+
+
+class VideoJobCallback(BaseModel):
+    job_id: int
+    status: str
+    frames: list[FrameInfo] | None = None
+    output_path: str | None = None
+    error_message: str | None = None
+
+
+class VideoAcceptedResponse(BaseModel):
+    message: str
+    job_id: int