Fix to avoid moving batch to device for DataParallel

[rohitgr7]

What does this PR do?

Discussed here: #10649 (comment)

Does your PR introduce any breaking changes? If yes, please list them.

Before submitting

• Was this discussed/approved via a GitHub issue? (not for typos and docs)
• Did you read the contributor guideline, Pull Request section?
• Did you make sure your PR does only one thing, instead of bundling different changes together?
• Did you make sure to update the documentation with your changes? (if necessary)
• Did you write any new necessary tests? (not for typos and docs)
• Did you verify new and existing tests pass locally with your changes?
• Did you list all the breaking changes introduced by this pull request?
• Did you update the CHANGELOG? (not for typos, docs, test updates, or internal minor changes/refactorings)

PR review

Anyone in the community is welcome to review the PR.
Before you start reviewing make sure you have read Review guidelines. In short, see the following bullet-list:

• Is this pull request ready for review? (if not, please submit in draft mode)
• Check that all items from Before submitting are resolved
• Make sure the title is self-explanatory and the description concisely explains the PR
• Add labels and milestones (and optionally projects) to the PR so it can be classified

Did you have fun?

Make sure you had fun coding 🙃

cc @Borda @justusschock @awaelchli @akihironitta @rohitgr7

[justusschock]

LGTM!

Just a side-note: without transfering to gpu0 first in a non-blocking way, this may actually be a bit slower, since depending on how gpus are connected, data transfer between gpus can be then transfer between cpu and gpu :)

However, I still think we should merge this fix, since otherwise users are restricted in their batchsize since actually the batches of both gpus need to fit to one gpu first :)