[AZ-180] Enhance setup and improve inference logging

- Added a new Cython extension for the engine factory to the setup configuration.
- Updated the inference module to include additional logging for video batch processing and annotation callbacks.
- Refactored test cases to standardize the detection endpoint responses and include channel IDs in headers for better event handling.

src/main.py

@@ -645,6 +645,7 @@ async def detect_video_upload(
                     content_hash, _MEDIA_STATUS_AI_PROCESSED,
                     token_mgr.get_valid_token(),
                 )
+            await asyncio.sleep(0.01)
             _enqueue(channel_id, DetectionEvent(
                 annotations=[],
                 mediaId=content_hash,
@@ -681,8 +682,7 @@ async def detect_media(
     config: Annotated[Optional[AIConfigDto], Body()] = None,
     user_id: str = Depends(require_auth),
 ):
-    existing = _active_detections.get(media_id)
-    if existing is not None and not existing.done():
+    if media_id in _active_detections:
         raise HTTPException(status_code=409, detail="Detection already in progress for this media")
 
     channel_id = request.headers.get("x-channel-id", "")
@@ -779,7 +779,7 @@ async def detect_media(
             )
             _enqueue(channel_id, error_event)
         finally:
-            _active_detections.pop(media_id, None)
+            loop.call_later(5.0, lambda: _active_detections.pop(media_id, None))
             loop.call_later(10.0, _cleanup_channel, channel_id)
 
     _active_detections[media_id] = asyncio.create_task(run_detection())