revert: Gradio 5.x + ZeroGPU @spaces.GPU + demo.launch (was working, 4.x had Starlette bug)

fix: add @spaces.GPU stub to satisfy ZeroGPU checker (hardware setting needed in UI)

fix: cpu-basic hardware + gr.mount_gradio_app + uvicorn (removes ZeroGPU dependency)

fix: Gradio 4 HighlightedText unhashable dict error by returning list of tuples to UI

fix: change flagging_mode to allow_flagging for Gradio 4.x compatibility

refactor: switch to CPU-only — remove ZeroGPU/spaces, consistent latency

perf: async endpoint + startup warmup to reduce ZeroGPU cold-start latency

fix: include score in entity response (was missing from highlighted list)

fix: remove add_middleware (cannot add after server start, HF proxy handles CORS)

fix: use demo.launch() for ZeroGPU compliance + attach REST routes to gr_app

fix: use @spaces.GPU directly — ZeroGPU scanner requires literal decorator

fix: add uvicorn.run() to keep server alive — was exiting with code 0

fix: make @spaces.GPU conditional + fix Pydantic v2 ConfigDict

fix: pin huggingface_hub<0.26.0 — spaces base image locks gradio 4.x which needs HfFolder

fix: explicitly pin gradio>=5.0.0 to upgrade system gradio used by spaces

fix: upgrade to Gradio 5 to resolve HfFolder import error

fix: add HF Spaces README config

feat: add FastAPI REST endpoint /api/detect alongside Gradio UI

Add test-gradio HF Space for codemix PII detector