Improve new_group example in the context of SyncBatchNorm (#48897)
Summary:
Closes https://github.com/pytorch/pytorch/issues/48804
Improves some documentation/example in SyncBN docs to clearly show that each rank must call into all `new_group()` calls for creating process subgroups, even if they are not going to be part of that particular subgroup.
We then pick the right group, i.e. the group that the rank is part of, and pass that into the SyncBN APIs.
Doc rendering:
<img width="786" alt="syncbn_update" src="https://user-images.githubusercontent.com/8039770/101271959-b211ab80-373c-11eb-8b6d-d56483fd9f5d.png">
Pull Request resolved: https://github.com/pytorch/pytorch/pull/48897
Reviewed By: zou3519
Differential Revision: D25493181
Pulled By: rohan-varma
fbshipit-source-id: a7e93fc8cc07ec7797e5dbc356f1c3877342cfa3