refactor: reorganize code and use c api (#133)

model.h

@@ -117,6 +117,9 @@ public:
     ggml_type get_sd_wtype();
     std::string load_merges();
     bool load_tensors(on_new_tensor_cb_t on_new_tensor_cb, ggml_backend_t backend);
+    bool load_tensors(std::map<std::string, struct ggml_tensor*>& tensors,
+                      ggml_backend_t backend,
+                      std::set<std::string> ignore_tensors = {});
     int64_t cal_mem_size(ggml_backend_t backend);
     ~ModelLoader() = default;
 };