This pull request introduces a new @orama/plugin-pgvector package, enabling Orama to use PostgreSQL with the pgvector extension as a backend for persistent, memory-efficient vector search and storage. The implementation offloads document storage and vector operations to PostgreSQL, supports lazy loading and memory limits, and provides a test script for quick validation.

The most important changes are:

New Plugin Implementation:

• Added the @orama/plugin-pgvector package with a package.json defining dependencies on pg, pgvector, TensorFlow libraries, and Orama, and including build/test scripts.
• Implemented the main plugin logic in src/index.ts, which:
  • Hooks into Orama's lifecycle to store documents in PostgreSQL, remove them from memory, and perform vector searches directly in the database.
  • Supports lazy loading, memory limits, and a vector-search-only mode for efficiency.
  • Manages document caching and memory usage with an LRU strategy.

Database Management and Vector Operations:

• Added src/pgvector-manager.ts to handle:
  • Table creation, schema adaptation, and vector column/index management in PostgreSQL.
  • Document insertion/removal, vector search queries, and memory cache management.
  • Efficient retrieval and reconstruction of documents with vector data.

Testing and Usage Example:

• Provided a comprehensive test script test/quick-test.js that:
  • Sets up a test database, inserts sample documents with mock embeddings, and performs vector similarity searches.
  • Demonstrates plugin usage and verifies correct integration with Orama and PostgreSQL.