Feature request: document/support llama-server HTTP endpoint for OpenAI-compatible serving

Background

setup_env.py already builds llama-server as part of the cmake step (it lives at build/bin/llama-server after a successful build). This binary provides a fully OpenAI-compatible HTTP API (/v1/chat/completions, /v1/completions, /v1/models) — the same interface as llama.cpp's server.

The README currently only documents run_inference.py for inference. The server binary is silently present but undiscovered by most users.

What this unlocks

• Drop-in replacement for OpenAI API in downstream tools (LangChain, Open WebUI, custom apps) without code changes
• Persistent model loading (no 2-3s cold-start per request)
• Integration with job queues or proxy layers that speak OpenAI protocol

Minimal usage (after build)

./build/bin/llama-server     --model models/BitNet-b1.58-2B-4T-gguf/ggml-model-i2_s.gguf     --host 127.0.0.1     --port 8080     --parallel 1     --ctx-size 4096

# Then:
curl http://127.0.0.1:8080/v1/chat/completions   -d '{"model":"bitnet","messages":[{"role":"user","content":"Hello"}]}'

Question

Would the team be interested in a PR that:

1. Documents this capability in the README (a single section — no code changes)
2. Optionally adds a minimal Python wrapper script (consistent with the repo's Python-first style) to make the invocation discoverable

Happy to contribute either or both if there's interest. Flagging as a question first rather than opening a cold PR.

Start with the README section covering run_inference.py and inspect setup_env.py to confirm where build/bin/llama-server is produced. Document the shown server command and OpenAI-compatible endpoints, and clarify whether the Python wrapper is in scope. Done means a newcomer can discover and launch the HTTP server from the README.