fix: improve face clustering accuracy with similarity threshold and p… (#771)

* fix: improve face clustering accuracy with similarity threshold and post-clustering merge

- Added similarity_threshold parameter (0.85) to filter dissimilar faces before clustering
- Implemented post-clustering merge to combine duplicate clusters of same person
- Added quality filtering for embeddings (std and norm checks)
- Improved distance calculation using precomputed cosine distances
- Enhanced logging for better debugging and monitoring
- Prevents side-face false positives while maintaining high accuracy

* coderabbit minor changes

* coderabbit review minor changes

* fix: add NaN/zero-vector guards and format code

- Added _validate_embedding() to check for NaN/inf/zero vectors
- Guard against invalid embeddings in clustering and assignment
- Safe division in cosine distance calculations
- Format code with Black for lint compliance

---------

Co-authored-by: Keshav Sharma <keshavsharmaog@Keshavs-MacBook-Air.local>

Author: keshaviscool

.gitignore

@@ -31,3 +31,4 @@ images_cache.txt
 videos_cache.txt
 venv/
 frontend/dist
+env/