Trigger Objects

Dispatch object detection and tracking on video chunks. Known issues: - Only detects people (via /person/track). The feature is named "objects" but other BoundingBoxType values (face, text, facecam) are never produced here; filtering get_objects by those types silently returns nothing. - person_id is always null in API responses. The tracker assigns integer tracking IDs per chunk, which are converted to UUID strings before DB storage. The API reader calls int() on those UUIDs, which always fails. - person_id is chunk-scoped, not video-scoped. The tracker resets IDs at each chunk boundary, so the same person gets a different person_id in every chunk — cross-chunk re-identification does not work.