Speed up CPU inference: halve token limits, pre-download models, fix OMP threads

Use bfloat16 on CPU to halve memory (8GB vs 16GB float32)

Fix MCP subprocess deadlock: use stderr=None instead of PIPE

Add timeout and stderr logging to MCP subprocess to debug tool hangs

Force MCP tool models to CPU to avoid GPU VRAM contention with MedGemma

Add RAG Phase 4 management guidance, rebuild guidelines index (286 chunks), post-analysis hint UI

Use dtype instead of deprecated torch_dtype in model_kwargs

Redesign chat UI and fix MedGemma generation config issues

Pass HF_TOKEN explicitly to pipeline() for gated model auth

Use HF_TOKEN env var to authenticate for gated MedGemma model

Initial commit — SkinProAI dermoscopic analysis platform