feat: add lightweight trajectory re-identification utility

[AnandMayank]

Description

What is this PR

• Bug fix
• Addition of a new feature
• Other

---

Why is this PR needed?

Current pose estimation pipelines (e.g. DLC, SLEAP) assume a fixed pool of individual IDs defined at training time. In ethology and field recordings, this assumption often breaks:

• animals leave and re-enter the frame,
• long occlusions occur,
• multiple similar-looking individuals appear sequentially.

This can result in identity reuse, where different individuals are incorrectly merged into a single ID, leading to fragmented or semantically incorrect trajectories and unreliable downstream analysis.

While gap-based splitting (e.g. split_trajectories_by_gap, see #801) provides a useful baseline, it is insufficient for long-horizon recordings, where:

• the same individual may reappear after long gaps, or
• different individuals may appear after short gaps.

---

What does this PR do?

This PR introduces a lightweight, optional re-identification utility in ethology that operates as a post-processing step on existing trajectory data:

• adds a new utility for trajectory re-identification based on simple motion consistency
• operates on gap-split trajectories (e.g. output of split_trajectories_by_gap)
• maintains a short-term trajectory memory (position and velocity statistics)
• attempts to re-associate new segments with previous trajectories when:
  • motion similarity is high, and
  • the implied motion is physically plausible
• assigns a new ID when no plausible match exists
• preserves all original coordinates, metadata, and attributes
• designed as an experimental ethology utility, not a replacement for full tracking systems

The association logic is inspired by boxMOT-style multi-object tracking, in the sense that:

• detection/trajectory generation is decoupled from identity association, and
• re-identification is handled via lightweight motion-based heuristics.

boxMOT itself is not introduced as a dependency — the implementation remains fully self-contained and Ethology-native.

---

References

• Related discussion in Movement: #792 (Support re-IDing data)
• Related Movement PR: #801 (trajectory gap splitting)
• boxMOT (conceptual inspiration for association logic)

---

How has this PR been tested?

Unit tests were run locally with pytest, including:

• basic re-association of split trajectories,
• correct handling of long vs short gaps,
• rejection of physically implausible matches,
• multiple individuals processed independently,
• edge cases (no candidates, empty memory),
• preservation of metadata and coordinates.

---

Is this a breaking change?

No.
This feature is optional and does not affect existing pipelines.

---

Does this PR require an update to the documentation?

Not strictly required.
A short usage example can be added in a follow-up PR once the API stabilizes.

---

Checklist

• The code has been tested locally
• Tests have been added to cover new functionality
• Documentation has been updated (optional follow-up)
• Code has been formatted with pre-commit