fix(deploy): force SentenceTransformer to CPU mode for ZeroGPU compatibility

fix(deploy): import torch and disable CUDA before importing spaces

fix(deploy): also patch _cuda_init during harness init, not just import

fix(deploy): temporarily patch _cuda_init during torch import to bypass ZeroGPU emulation

fix(deploy): disable CUDA before torch import to fix ZeroGPU emulation conflict

fix(deploy): run init at module level - gradio.Server ignores on_event startup

feat(deploy): add /status endpoint for remote diagnostics

fix(deploy): non-blocking index build so app starts immediately

fix(deploy): improve startup diagnostics and index build detection

fix(deploy): add detailed startup logging and subprocess error capture

fix(deploy): build index in startup event since HF Spaces ignores CMD

fix(deploy): add @spaces.GPU placeholder for ZeroGPU compatibility

fix(deploy): use gradio.Server() without args - it IS the FastAPI app

feat(deploy): add Hugging Face Spaces support via Gradio SDK