[FEATURE] Add pgvector support for scalable vector similarity search

[kishan0725]

Problem

The current PostgreSQL vector storage implementation uses in-memory cosine similarity computation in JavaScript, which doesn't scale for multi-user production deployments.

Performance impact:

• With 100K vectors: ~500ms query latency (unacceptable)
• With 1M vectors: ~5s latency + memory pressure
• Network overhead: Transfers ALL vectors over wire
• CPU overhead: JavaScript cosine similarity (single-threaded)

Proposed Solution

Leverage pgvector extension for native PostgreSQL vector operations.

Expected improvements:

• 100x faster queries (50ms vs 5000ms)
• Scales to millions of vectors
• Lower memory usage
• Supports approximate nearest neighbor (HNSW)

Code of Conduct

• I agree to follow this project's Code of Conduct