pi: fix llama.cpp provider discovery with auth

Add api, authHeader, and discovery.type fields so omp can
discover models via GET /v1/models with the Bearer token.

home-manager/progs/pi.nix

@@ -28,6 +28,9 @@ let
       "llama.cpp" = {
         baseUrl = "https://llm.sigkill.computer";
         apiKey = lib.strings.trim (builtins.readFile ../secrets/llama_cpp_api_key);
+        api = "openai-responses";
+        authHeader = true;
+        discovery.type = "llama.cpp";
       };
     };
   };