Require API key in Gradio UI before model inference

Fix Gradio 6.x chatbot history format (tuples → message dicts)

Fix NaN/inf sampling error: switch CPU-cached models to bfloat16

Call hf_login() at startup to authenticate gated model access

Add X-API-Key auth to HF Space inference endpoint

Fix ZeroGPU startup: use demo.launch() instead of uvicorn directly

Drop fastapi/httpx/huggingface-hub from Space requirements

Remove gradio/spaces/uvicorn from Space requirements.txt

Switch HF Space SDK from docker to gradio to enable ZeroGPU

Add HF ZeroGPU Space and migrate backend to huggingface_space provider