Add measured-latency notice for Space + accurate timing in README

More CPU headroom — 900s, 160 tokens, real-shaped JSON warmup

Tune for CPU-only HF Space — 8min timeout, 200 tokens, keep_alive 24h

app.py — bind via GRADIO_SERVER_NAME env (default 127.0.0.1)

Add zstd — Ollama installer requires it for bundle extraction

Initial Space: Docker + Ollama + Gradio + WHO IMCI

initial commit