[Functional Collectives] Migrate DeviceMesh::all_reduce to use functional all_reduce. (#95009)
BC: This changes the signature and semantics of DeviceMesh::all_reduce.
DeviceMesh::all_reduce now uses a functional collective under the hood which makes it more easily traceable.
You no longer need to use CommTensor to get a trace.
all_reduce now is async only and uses AsyncCollectiveTensor to ensure proper stream synchronization.
Signature changed: removed `async_op` param and changes return type from `Optional[Work]` to `torch.Tensor`.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/95009
Approved by: https://github.com/wanchaol
