move torch.cuda.set_device() to enable collective calls earlier in setup

[awaelchli]

What does this PR do?

Unblocks #8017 who is calling trainer.log_dir early in Callback.setup(), which internally leads to a broacast, but hangs because device is not yet set!

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 update the CHANGELOG? (not for typos, docs, test updates, or internal minor changes/refactorings)

PR review

Anyone in the community is free to review the PR once the tests have passed.
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?

I made sure I had fun coding 🙃

[codecov]

Codecov Report

Merging #8312 (3f9a9e2) into master (f796cb4) will decrease coverage by 0%.
The diff coverage is 80%.

@@          Coverage Diff           @@
##           master   #8312   +/-   ##
======================================
- Coverage      92%     92%   -0%     
======================================
  Files         213     213           
  Lines       13798   13793    -5     
======================================
- Hits        12744   12726   -18     
- Misses       1054    1067   +13     

[tchaton]

LGTM !

[carmocca]

Do we also want to move?:

pytorch_lightning//plugins/training_type/horovod.py:136:            torch.cuda.set_device(self.root_device)
pytorch_lightning//plugins/training_type/ddp_spawn.py:333:            torch.cuda.set_device(self.root_device)

[awaelchli]

@carmocca the one in ddp_spawn cannot be moved, it needs to set the device in the spawned subprocesses too.
for horovod I'm not confident enough to change it right now, as we have very little testing around that plugin.