llama: only use one iGPU device by default (#23897)

src/llama.cpp

@@ -225,7 +225,9 @@ static bool llama_prepare_model_devices(const llama_model_params & params, llama
                     }
 
                     case GGML_BACKEND_DEVICE_TYPE_IGPU:
-                        igpus.push_back({false, dev});
+                        if (igpus.empty()) {
+                            igpus.push_back({false, dev});
+                        }
                         break;
                     case GGML_BACKEND_DEVICE_TYPE_META:
                         GGML_ABORT("fatal error");