perf: add hnsw vector index for pgvector

[ademboukabes]

Description
Adds HNSW (Hierarchical Navigable Small World) vector indexes to the users and photo_faces tables for the pgvector embeddings.

Reasoning
Currently, queries executing face_embedding <=> $1 (Cosine Distance) are performing full sequential scans across the entire table. As the dataset of uploaded photos and user faces grows, this will result in O(N) search complexity, causing severe CPU spikes and degraded API latency.

Adding an HNSW index with vector_cosine_ops reduces the search complexity to O(log N), allowing the AI matching pipeline to scale efficiently with minimal performance overhead.

Changes

• Created Alembic migration 9ec59aeb192b_add_hnsw_indexes
• Added CREATE INDEX using hnsw on users.face_embedding
• Added CREATE INDEX using hnsw on photo_faces.embedding