refactor(detector): replace monolithic detector with pluggable multi-backend package

feat(ui): add manual bbox annotation fallback when detection fails

feat(ui): add capture mode selector (single/multi/video WIP)

refactor(vision): use detector crops for per-garment extraction

feat(vision): add verbose description to garment extraction

feat(catalog): add image persistence per garment

feat(wardrobe): implement vision pipeline, catalog and assistant

feat (wardrobe): prepare project structure and basic gradio ui