fix(deploy): wrap hf_backend startup gpu_warmup in try-except block to handle transient HF GPU hardware ECC errors

fix(deploy): run gpu_warmup() at the top-level during container startup to satisfy ZeroGPU check

fix(deploy): add torch dependency and PyTorch CUDA tensor allocation inside @spaces.GPU to satisfy ZeroGPU PyTorch runtime checks

fix(deploy): register @spaces.GPU function on blocks.load to pass ZeroGPU event graph inspection

fix(deploy): declare @spaces.GPU function at the top level to pass Hugging Face ZeroGPU AST check

fix(deploy): add dummy @spaces.GPU function to pass Hugging Face ZeroGPU startup check

fix(deploy): run uvicorn directly in _launch_with_app to bypass Gradio 5.x launch() app argument deprecation

fix(deploy): pass title and theme to gr.Blocks constructor instead of blocks.launch to resolve TypeError in Gradio 4.x

fix(deploy): add demo.launch() to hf_backend/app.py so HuggingFace Space Gradio SDK doesn't exit immediately with code 0

fix: inject combined FastAPI app via launch(app=combined) instead of separate uvicorn.run, fix Gradio 6 title/theme warnings

fix: run uvicorn at module level in hf_backend/app.py instead of monkey-patching blocks.launch

Add HF Space Gradio SDK backend (free tier)