Enhance ObjectDetector: adjust identity similarity and add cross-identity merge threshold for improved face grouping

Enhance face segment building: improve clip creation logic and add identity matching based on embedding similarity

Enhance ObjectDetector: track face occurrences and add quality metrics for improved detection accuracy

Enhance ObjectDetector: add quality filters for face detection based on area, sharpness, and quality score

Enhance face detection: add identity grouping and similarity threshold for improved accuracy

Enhance face detection UI: add progress indicators and callback for real-time updates

Enhance face detection capabilities: add ObjectDetector class, model configuration, and ensure InsightFace weights management

Rebuild as music-only highlight extractor: Whisper + Librosa, remove multi-modal stack

Initialize ShortSmith v4 - scene-anchored clip boundaries, bug fixes from v3