Accommodate FSDP full-precision param_dtype training with PyTorch < 2.0

[speediedan]

Fixes #18277

When FSDP training with full precision param_dtypes (16-mixed, bf16-mixed and 32-true configurations) and PyTorch < 2.0, FSDP training will encounter this assertion error.

This is because FSDP uses the noneness of param_dtype as a proxy for the _uses_param_mixed_precision property and FSDPPrecisionPlugin currently sets the default param_dtype to torch.float32 when training in full precision.

This PR:

1. Sets MixedPrecision param_dtype to None when FSDP training with full precision param_dtypes and PyTorch < 2.0
2. Updates relevant Lightning and Fabric tests

Before submitting

• Was this discussed/agreed 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 minor internal changes/refactors)

PR review

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

Reviewer checklist

• 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

[speediedan]

I believe the currently failing check (Test Fabric / fabric-cpu (ubuntu-20.04, lightning, 3.8, 1.11): test_collectives_distributed[2]) is not related to this PR.