Better error message when trying to re-initialize CUDA in forked subprocess

[awaelchli]

What does this PR do?

Follow up to #14631
Follow up to #13405

We recently launched DDP Fork support in notebooks (ddp_notebook), and some users will soon report issues when seeing this error message in their notebooks: "Cannot re-initialize CUDA in forked subprocess".

The issue is mitigated by #14631 for our internal calls, and pytorch addresses torch.cuda utility function calls in pytorch/pytorch#84879 and pytorch/pytorch#85024. However, it can always happen that the user allocates CUDA memory accidentally without knowing that this is not supported with the fork start method. We point it out in our documentation, but that's not enough.

This PR introduces an error message that is more helpful than the default PyTorch one.

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

No

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 @carmocca @justusschock @awaelchli @akihironitta

[carmocca]

I am subscribed to your PRs, reviewing them before you even mark them as ready 🐎

[awaelchli]

I am subscribed to your PRs, reviewing them before you even mark them as ready 🐎

Thank you! Is that a horse?

Will mark it ready once I gave this a final check in Google Colab as well.

[akihironitta]

LGTM!