Pass HF_TOKEN explicitly to SentenceTransformer for gated model download

Optimize PyTorch threading, run generation in thread, and pre-warm model at startup

Optimize referee model to Qwen2.5-0.5B-Instruct for CPU and add warning popup

Configure for 100% self-contained CPU deployment

RAG retreival updates

Restore Phase 5 advanced RAG changes and clean schemas.py

Merge branch 'main' into feature/advanced-rag

Pushing the changes for judge as LLM

changes for readme file and UI changes

local: phase 5 arena and text highlighter