llama: end-to-end tests (#19802)

* tests: add end-to-end tests per model architecture

* fixup for rebase

* fix use-after-free in llama-model-loader.cpp

* fix CI

* fix WebGPU

* fix CI

* disable CI for macOS-latest-cmake-arm64

* use expert_weights_scale only if != 0.0f

* comments

src/models/baichuan.cpp

@@ -56,6 +56,7 @@ llm_build_baichuan::llm_build_baichuan(const llama_model & model, const llm_grap
                             );
                     break;
                 case LLM_TYPE_13B:
+                case LLM_TYPE_UNKNOWN:
                     break;
                 default:
                     GGML_ABORT("fatal error");