feat: load OCTRON bounding box CSVs (#1108)

[anshurajbisoyi98-ctrl]

Fixes #1108

OCTRON exports one CSV per tracked ID, with video metadata at the top of each file. This adds an OCTRON loader to the existing bounding-box loading interface and validates the metadata and tracking columns before constructing a dataset.

A single CSV loads its track. To combine tracks from the same video, pass their files with additional_files; the loader checks that the video name, frame counts, and dimensions match, and rejects duplicate frame/track observations. It preserves the original zero-based frame indices and missing detections, derives box centres and widths/heights from the corner coordinates, and uses OCTRON track IDs as individual names.

The user guide now documents OCTRON’s format and both loading patterns. The explicit additional_files interface is included in this draft for maintainer feedback.

Validation

• OCTRON and related loader/validator tests: 203 passed, 4 deselected.
• All pre-commit hooks passed.
• Sphinx HTML build passed with gallery execution disabled. The default gallery build timed out while fetching unrelated sample video data.
• GitHub Actions Tests and Docs runs currently require maintainer approval; no remote test result is available yet.

AI assistance: Codex assisted with implementation, regression tests, documentation, and verification.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud