Use _process_group in ReplicatedTensor and ShardedTensor.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/77209
As per comment in
https://github.com/pytorch/pytorch/pull/77191#discussion_r869641344, making
everthing consistent amongst these tensors.
Differential Revision: [D36300798](https://our.internmc.facebook.com/intern/diff/D36300798/)
Approved by: https://github.com/wanchaol, https://github.com/fduwjj
