Add support for CPU AMP autocast

[SeanNaren]

What does this PR do?

Adds PyTorch 1.10 support for CPU autocast. Requires installing 1.10 (and I noticed that for now till the release, we have to include a dev check to the availability for people using nightly or building from source).

Now in 1.10 this works:

Trainer(precision='bf16')

cc @ananthsub @anj-s @min-xu-ai I noticed that torch.cpu.autocast only has bf16 support which means it might not play well with fp16 mode for FSDP when using CPU offloading. Would be nice to have support for this out the box!

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 list all the breaking changes introduced by this pull request?
• Did you update the CHANGELOG? (not for typos, docs, test updates, or internal minor changes/refactorings)

PR review

Anyone in the community is welcome to review the PR.
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?

Make sure you had fun coding 🙃

[codecov]

Codecov Report

Merging #9084 (4677e77) into master (e9f4bff) will increase coverage by 0%.
The diff coverage is 94%.

@@          Coverage Diff           @@
##           master   #9084   +/-   ##
======================================
  Coverage      92%     92%           
======================================
  Files         176     176           
  Lines       14663   14666    +3     
======================================
+ Hits        13496   13500    +4     
+ Misses       1167    1166    -1     

[tchaton]

Dope !

[ananthsub]

very nice!

[daniellepintz]

@SeanNaren I think this PR caused the following 5 tests to fail in our CI job (conda 3.7, 1.10)

FAILED tests/models/test_amp.py::test_amp_cpus[1-bf16-None] 
FAILED tests/models/test_amp.py::test_amp_cpus[1-bf16-ddp_spawn] 
FAILED tests/models/test_amp.py::test_amp_cpus[2-bf16-None] 
FAILED tests/models/test_amp.py::test_amp_cpus[2-bf16-ddp_spawn] 
FAILED tests/models/test_amp.py::test_cpu_model_with_amp 

The issue tracking this is #9280

Stack traces for the failing tests can be found in this gist: https://gist.github.com/daniellepintz/3d5a13b6264a832fc745f73460d2fa23 (from https://github.com/PyTorchLightning/pytorch-lightning/runs/3483032531)

Do you know what might be causing these failures?