Always set detection age even when not kept in past_detections.

[cpruijsen]

Always set detection.age in _conditionally_add_to_past_detections, including when the sample is skipped and when past_detections_length is 0.

Fixes #317.

Detection.age was only written when a match was stored in
TrackedObject.past_detections. hit() still assigns that same detection to
last_detection, so once the buffer was full (or past_detections_length
was 0) last_detection.age stayed None.

The reporter (nmichlo) asked that age be set even when the detection is
discarded, "because this affects the last_detection."

What I chose and the alternative

The keep path already overwrote age in _conditionally_add_to_past_detections,
which is the site the reporter pointed at.

Alternative: set age in hit() only, so ReID merge() would not rewrite ages
of detections it does not keep. Also considered asserting age is unset.
Did not add that assert: merge() re-adds detections that already have age.

Can move the stamp to hit() instead.

Test plan

• tests/test_tracker.py::test_detection_age_always_set fails without
  the fix and passes with it (past_detections_length 0, 2, and 4)
• tests/ (unit tests) pass

[cpruijsen]

Going to close this. Not planning to keep it rebased.