Support fused Adam with mixed precision

[awaelchli]

What does this PR do?

Fixes #15501

PyTorch 1.13 introduced Adam(fused=True|False). It fuses several operations in the step() method of the optimizer to speed up execution (reduced number of kernel launches). However, since certain operations get fused together, we can no longer unscale the gradients independently in case we want to do gradient clipping. With a fused optimizer like Adam, the unscaling gets handled directly in the step method. Furthermore, a call to scaler.unscale_(optimizer) in our precision plugin currently leads to gradients being unscaled twice in the case of fused Adam, which is undesired. We address both issues in this PR.

A more detailed explanation is here.

In summary:

• If you are using precision=16 AND fused Adam, this PR fixes optimization parity with non-fused Adam
• If you are using precision=16 AND fused Adam AND also gradient clipping, there will be an error saying it is not supported.

Follow up work:

• Currently, the added parity test is skipped in GPU CI. We need to upgrade CI to 1.13.

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

None intended.

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