LLM inference (Llama-3.1-8B via Cloudflare Workers AI). POST { prompt, system? } → { response, model }.