feat(analysis): add shot content understanding

[TristinWen]

Summary

• generate one editing-oriented .semantic.json containing required media facts, contiguous shots, representative timestamps, and structured scene descriptions
• detect cuts and extract representative frames with FFmpeg while keeping model providers replaceable
• add OpenAI structured vision and reviewed JSON import providers
• make technical details, embedded metadata, SHA-256, and raw FFprobe data opt-in through repeatable --include
• expand scene semantics, JSON Schema, CLI documentation, roadmap, and changelog

Validation

• uv run ruff check .
• uv run mypy src tests
• uv run pytest -q --basetemp <writable-directory> — 76 passed, 96.91% coverage
• processed D:\BaiduNetdiskDownload\20240610-3.MP4 end to end into a schema-valid 3-shot semantic manifest

Privacy

• the source video, extracted representative frames, and its reviewed descriptions are not included in this pull request