Cross-OS federated gateway unifying every OpenAI-compatible inference backend (Cognis fleet, Ollama, llama.cpp, vLLM) behind one /v1 endpoint