Implement full GPU/Rank monitoring dashboard

Use chat_completion with Mistral v0.2

Use text_generation API instead of chat_completion

Switch to Mistral 7B Instruct (popular free model)

Switch to Gemma 2 2B model

Use huggingface_hub InferenceClient for reliable API access

Require HF_TOKEN for inference API authentication

Fix HF Inference API endpoint (use router.huggingface.co)

Add HuggingFace Inference API support for HF Spaces

Update dashboard to connect to real vLLM server

Simplify app - remove timer, make all tabs work

Fix demo mode with simulated metrics

Initial commit: LLM Inference Dashboard