Implement Vector Storage with LanceDB and Transformers.js

[prosdev]

Description

Implement vector storage and retrieval system using LanceDB for embedded storage and @xenova/transformers for local embedding generation. This replaces the original plan for Chroma DB to better support a local-first, serverless architecture.

Acceptance Criteria

• Vector storage system initializes using LanceDB (embedded)
• Embedding generation uses @xenova/transformers (all-MiniLM-L6-v2)
• System downloads and caches the embedding model on first run
• Vector search returns relevant results ranked by cosine similarity
• Metadata is stored efficiently alongside vectors
• System handles standard repository sizes efficiently without a separate server process

Technical Requirements

• Integrate @lancedb/lancedb for serverless vector storage
• Integrate @xenova/transformers for local embedding generation
• Implement efficient batching for embedding generation
• Define clear interfaces for 'VectorStore' and 'Embedder'
• Ensure cross-platform compatibility for the native bindings
• Add unit tests for storage and retrieval operations

Branch: feat/vector-storage
Priority: High
Estimate: 4 days
Parent Epic: #1