[Database] Implement FAISS vector store wrapper + hybrid metadata retrieval

[saviornt]

Description

Create a clean, persistent FAISS vector store abstraction that links vectors to SQLite metadata (incidents / model versions) for efficient RAG-style retrieval in the Assistant (and limited edge use).

Type

• Task

Focus Area (pick one)

• Shared Utils & Models

Priority

• Medium

Acceptance Criteria

• File: shared/src/database/faiss.py — low-level wrapper
• File: shared/src/database/faiss_service.py — higher-level service
• Supports:
  • add_vectors(vectors: np.ndarray, metadata: list[dict]) → list[int]
  • search(query_vector: np.ndarray, k: int = 5) → list[dict] (with score + metadata)
• Default index: IndexFlatL2 (MVP); future path for IndexIVFFlat
• Persistent save/load via faiss.write_index / faiss.read_index
• Index path from Pydantic Settings
• Metadata rows inserted into faiss_vectors_metadata table (vector_id FK)
• Async-friendly (run FAISS sync calls in thread if needed)
• Health check: index size, dimension consistency
• Unit tests: mock vectors, search recall, persistence round-trip
• Google-style docstrings + full type hints

Blocker / Dependencies

• [Database] Define Pydantic v2 models and initial relational schema
• [Database] Implement robust async SQLite session factory + connection pooling

Notes / Links

• Related files: shared/src/database/schemas.py, shared/src/core/settings.py
• Reference: FAISS wiki + Python bindings documentation