feat: Remove LLM dependencies and integrate HNSW-based ANN for faster/stable search ...

- Removed llama-server, GGUF models, and heavy LLM DLLs.
- Cleaned up build.rs and tauri.conf.json to lightweight the build.
- Integrated hnsw_rs in mcp.rs for high-performance semantic search.
- Improved LSA robustness for unknown words.
- Reclaimed approx 3GB of workspace space.