Default to precision=bf16 on CPU when precision=16 is passed

[carmocca]

What does this PR do?

With this change, Trainer(precision=16, accelerator='cpu) now runs with precision='bf16' automatically.
This is desirable for no-code-change transitions between environments with different accelerators, for example, local (CPU only) to Colab (GPU or TPU)

Trainer(amp_backend="apex", precision=16, accelerator='cpu') will still raise an error as apex does not support bf16

Part of #10027

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

None

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?
• [n/a] 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

• 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

[awaelchli]

nice

will you add a changelog entry? because this would be a backward incompatible change since the default type changes

[codecov]

Codecov Report

Merging #10033 (9fbb921) into master (d45897d) will increase coverage by 0%.
The diff coverage is 90%.

@@           Coverage Diff           @@
##           master   #10033   +/-   ##
=======================================
  Coverage      93%      93%           
=======================================
  Files         180      180           
  Lines       15870    15890   +20     
=======================================
+ Hits        14689    14720   +31     
+ Misses       1181     1170   -11     

[tchaton]

LGTM !

[rohitgr7]

I think we support Trainer(precision="16"), do we?

[carmocca]

I don't think so because we don't convert the value to a PrecisionType (which works for int and str)

IMO we should, but should be done in a follow-up. The previous code here had the same check