perf: --eager-load to pre-load params at model-load time (#1687)

src/model_manager.h

@@ -158,6 +158,7 @@ public:
     }
 
     bool validate_registered_tensors();
+    bool load_all_params_eagerly();
 
     bool prepare_params(const std::vector<ggml_tensor*>& tensors) override;
     void release_compute_backend_params(const std::vector<ggml_tensor*>& tensors) override;