Fix bug where __getstate__ of DDP looks for self._replicated_tensor_module
Pull Request resolved: https://github.com/pytorch/pytorch/pull/76349
When we are not using ReplicatedTensor in DDP and try to save a DDP
module it will error out since it tries to delete the _replicated_tensor_module
attribute.
Fixing this by checking if this mode is enabled before triggering the delete.
Differential Revision: [D35875167](https://our.internmc.facebook.com/intern/diff/D35875167/)
Approved by: https://github.com/mrshenli, https://github.com/zhaojuanmao
