Associate or separate IDs after single-camera pipeline runs

[mattdawkins]

Single-camera pipeline runs on stereo or multicamera datasets can produce track IDs that collide with existing annotations on another camera. When another camera has tracks, DIVE now asks whether to associate the results after the pipeline finishes.

• No: Assign every output track an ID above the maximum ID on all other cameras, keeping repeated states of a track together.
• Yes: Require stereo mode, enabled stereo features, and a loaded calibration file before launching. Multicamera association reports that support is not implemented. After detection, run VIAME's calibration-based stereo pairing and import the associated results for both cameras.
• Cancel: Do not launch the pipeline.

The viewer saves pending annotations before checking other cameras. Desktop and server jobs process results before importing them, preserve original IDs when other cameras are empty, and fall back to detector output when the track file contains only a header. Association failures do not import either camera's results. The VIAME job uses accumulated track pairing so matches discovered later can also update earlier track states.

Validation on DIVE main:

• 176 targeted client tests passed, including annotation ingestion, camera handling, association decisions, and result processing.
• Type checking passed.
• Changed client files passed ESLint before moving to main; the feature transferred without conflicts.
• Four Python tests passed, including a real run of the installed VIAME executable verifying matched IDs, unmatched detections, late track pairing, and unchanged frame numbers.
• Full Girder integration was not run because worker dependencies are unavailable locally.

Uses the existing VIAME compute_measurements process with calibration-based detection pairing and accumulated track output; no VIAME source changes are included.