Validate that tensors are contiguous in ProcessGroupNCCL
Fixes https://github.com/pytorch/pytorch/issues/77554 by ensuring we
require contiguous tensors for send/recv in ProcessGroupNCCL.
Differential Revision: [D36500769](https://our.internmc.facebook.com/intern/diff/D36500769/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/77809
Approved by: https://github.com/rohan-varma, https://github.com/wanchaol
