Lite: Support self.log from a LightningModule

[lightningforever]

What does this PR do?

Enables LightningModule users to call self.log() and self.log_dict() in any hook when used together with Fabric.

• We support logging Python scalars and Tensor scalars.
• There is basic input validation for the right data types and nesting.
• You can't pass in a torchmetrics.Metric object. You need to call .compute() yourself at the right time. This is not a limitation, this is a feature. Since compute() is expensive, Fabric users need to control themselves when to do it.
• We don't support ANY of the optional arguments of self.log namely: prog_bar, on_step, on_epoch, reduce_fx, enable_graph, add_dataloader_idx, batch_size, sync_dist, sync_dist_fn, sync_dist_group, metric_attribute, rank_zero_only. Supporting them would require making assumptions about loop structure and features that do not exist, and will never exist in Fabric. We silently ignore them.

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 🙃

[justusschock]

LGTM in general. One thing: should we warn the user the first time they pass in unsupported arguments? Otherwise, this may silently lead to unexpected behavior

[awaelchli]

One thing: should we warn the user the first time they pass in unsupported arguments? Otherwise, this may silently lead to unexpected behavior

I initially wanted to do that. But it would add quite some code to do this correctly and nicely, with added complexity. As a consequence, all types and default values would change in the signature to Optional[None] = None etc. I didn't want to go down that path without knowing this is actually worth it.