Add reusable video preparation and raw measurement APIs

[kstonekuan]

Video integrations currently have to rebuild camera export, model-video preparation, and raw-measurement helpers outside HFlow. Add public file-level APIs so callers can reuse those operations without creating an episode or persistent workspace.

• Export a selected MCAP camera while retaining irregular timestamps and decoded pixels; reject unsupported media and never overwrite existing output.
• Add nearest-position keyframe sampling, unpadded fitting, and explicit JPEG quality while preserving current sampling defaults.
• Expose frame statistics with explicit luma-range and cache policy, raw blur measurements, and duration-weighted camera-shake summaries.
• Prepare model video through the existing JPEG and canonical H.264 recipe without an intermediate MCAP. Existing transform bytes/defaults are unchanged.
• Document the public contracts and cover timestamps, pixel parity, coverage, invalid inputs, and output preservation with synthetic media.

Validation:

• uv run --locked ruff check --fix
• uv run --locked ruff format
• uv run --locked ty check
• uv run --locked pytest -q: 2,254 passed, 8 skipped
• Python 3.11 affected API tests: 105 passed
• Downstream HUD media/pipeline regressions: 178 passed
• Lychee checks for changed documentation: 55 links passed
• Python 3.12 native overlay: all seven changed implementation modules compiled; all 105 affected API tests passed against the compiled modules