API: tigger motion event should support object an tags

[titro]

The trigger motion event is fine but somehow useless is you can not send more information in the request what causes the motion trigger.

Would be wonderful if you could send the object(s) and tags with the request for motion event to add the eents on the timeline of you already have 7/24 recording.

[matteius]

Done in 0.36.5. POST /api/motion/trigger now takes the object classes along with the event:

curl -X POST http://lightnvr:8080/api/motion/trigger \
  -H "X-API-Key: <key>" -H "Content-Type: application/json" \
  -d '{"stream":"Driveway","action":"pulse","duration_ms":10000,
       "objects":[{"label":"person","confidence":0.93}],
       "tags":["frigate","front-gate"]}'

• label / objects — stored as detections, so they appear on the timeline, come back from the detections API, and go out over MQTT with the same shape as model detections (Home Assistant can filter them with selectattr('label','eq','person') exactly like ONVIF smart events). objects accepts bare strings (["person","vehicle"]) or per-entry confidences, so you can forward detector output verbatim. Use the same vocabulary the detectors use: person, vehicle, bicycle, face, animal, motion.
• tags — applied to the recording currently open for that stream.

There was also a blocker for exactly the case you described. The endpoint used to return 409 when the stream didn't have detection-based recording enabled — so with 24/7 recording there was nothing to trigger and you could never annotate the timeline at all. That's gone: the request is accepted, the event is recorded, and the response reports recording_triggered: false so you can still tell the two situations apart.

Two things worth knowing:

• Detections are recorded on the leading edge only (start and pulse) — a stop says nothing new about what was seen. They're passed through the stream's zone filter like model detections, so detections_stored in the response can be lower than what you sent.
• Tags attach to the currently open recording. With 24/7 recording that's the segment your event falls inside, which is what you want. With detection-based recording no segment exists yet at the instant the trigger arrives, so tags_applied comes back 0.

Full docs: docs/API.md → Trigger Motion Event

The 0.36.5 image is building now.