refactor(mge/tensor): support scalar in detach and grad

GitOrigin-RevId: 38349c8357871d853219026d1431b98ca55fe3a4

imperative/python/src/tensor.cpp

@@ -495,9 +495,10 @@ PyObject* TensorWrapper::detach() {
         new_tensor = std::make_shared<Tensor>(m_tensor->m_var);
     }
     new_tensor->m_trace_info = m_tensor->m_trace_info;
+
+    new_tensor->m_flags = m_tensor->m_flags;
     auto ret = TensorWrapper::make(pytype, std::move(new_tensor));
     return ret.release().ptr();
-
 }
 
 PyObject* TensorWrapper::_dev_tensor(){