Fix inference: switch to OpenAI-compatible /v1/chat/completions endpoint

Add README.md with HuggingFace Spaces configuration frontmatter

Initial setup: React + Express streaming chat for Qwen 2.5 7B