ui: improve Gradio demo layout and fix inference time warning

fix: add threading lock around LLM inference to prevent concurrent call corruption

Add Gradio demo UI to HF Space (mounted at root alongside FastAPI)

fix(hf-space): switch to llama-cpp-python + GGUF to fix CPU OOM

fix(hf-space): float16+low_cpu_mem, hf-transfer, peft 0.18.1, retry logic

feat(p3-integration): remote-mode extractor + HuggingFace Space deployment