Fix device parser logic to avoid creating CUDA context

[awaelchli]

What does this PR do?

Fixes #14292

Discussed several solutions offline with @carmocca and @justusschock:

• Rearrange logic in AcceleratorConnector: not feasible
• Use a different start method to launch the process in which we evaluate the cuda function: too costly
• Detect whether we are running inside of ray process: Not possible

Opened issue on torch side: pytorch/pytorch#83973

The current solution seems to be the only one that allows us to have support for ddp-fork and at the same time be compatible with other frameworks. Ray users will have to set PL_DISABLE_FORK=1 for now.

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 🙃

cc @Borda @tchaton @rohitgr7 @justusschock @kaushikb11 @awaelchli @akihironitta