BrainForge/Security: direct vLLM (4090), HANA persona cache, Security stub

Add HANA klatchat credentials for BrainForge Security (4090 x1-3)

Add Mistral.ai as direct provider with Mistral Small 3.2, Small 4, and Devstral2

Simplify conversation ending: two clean closing messages after winding_down

Strip thinking content from all LLM responses and skip redundant role generation

Add AI-completed vs exact role generation style toggle

Improve freeform role generation prompt and expose finish_reason

Add response timing, chat stats, and role prompt viewer to settings

Rebrand UI with Neon.ai header, sidebar styling, expert personas, and UX improvements

Add speed-priority setting with 5s racing fallback for faster responses

Add Neon HANA routing, per-request timeouts, and chat brevity prompts

Add HuggingFace Spaces deployment with OAuth and rate limiting

Add freeform persona input mode with file upload and Developer menu toggle

Add Developer menu with orchestrator picker, download buttons, and LLMChats2-style dropdown

Initial commit: LLMChats3 - LLM conversation app with React + FastAPI, persona config, orchestrated conversation lifecycle, collapsible sidebar accordions, stop button, and export capabilities