[fix bug] add weight quant for combined model after optimized (#2929) (#2931)

* add weight quant for combined model after optimized, test=develop

lite/api/light_api.cc

@@ -25,6 +25,8 @@ void LightPredictor::Build(const std::string& lite_model_file,
   } else {
     LoadModelNaiveFromFile(lite_model_file, scope_.get(), &cpp_program_desc_);
   }
+
+  DequantizeWeight();
   BuildRuntimeProgram(cpp_program_desc_);
   PrepareFeedFetch();
 }