pub fn detect_backend(endpoint: &str) -> Result<String>Expand description
Blocking probe of /v1/models to detect which backend engine is running.
Inspects response headers (Server) and body content for tell-tale
strings from llama.cpp, SGLang, or vLLM. Returns a lower-case label
like "llama.cpp", "sglang", "vllm", or "unknown".
This is a synchronous helper so it can be called from the bench-harness
runner (which runs inside spawn_blocking).