Versioning of last checkpoints

[otaj]

What does this PR do?

Versioning of last checkpoints
Fixes #5030

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

The change should not be breaking if you do not rely on explicit name of the last checkpoint and use last_model_path instead.
One existing test needed to be adjusted.

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 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:

• 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 🙃

cc @Borda @carmocca @awaelchli @ninginthecloud @jjenniferdai @rohitgr7

[ananthsub]

This is a storage regression. For instance, this doubles the number of checkpoint files created when used with saving at the end of every epoch and setting save_last=True. Before, we'd have N+1 checkpoints, and now we have 2N checkpoints. The old "formerly last" checkpoints are still going to be kept around on disk.

This becomes noticeable for larger models.

The only way around this is within Lightning is to extend the ModelCheckpoint callback here and override all the hooks in order to be more performant.

[otaj]

This should save only one last checkpoint for a full run as long as the ModelCheckpoint callback's last_model_path does not change (which it shouldn't at the end of each epoch). Can you please show a repro so that we can fix the regression?