GGUF Backend Wiring Use a local llama.cpp or llamafile HTTP server, then run make backend-health and make gate.