LRs updates are called at the end of a skipped epoch

[LTMeyer]

What does this PR do?

This PR fixes learning rate not being updated at the end of epoch when on_train_batch_start returns -1.

Fixes #21296

It postpones the existing raise StopIteration after the learning rate update.

Before submitting

• Was this discussed/agreed via a GitHub issue? (not for typos and docs): Related issue LR is not updated when on_train_batch_start returns -1 #21296 hasn't been discussed yet.
• Did you read the contributor guideline, Pull Request section? Yes
• 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) No
• Did you write any new necessary tests? (not for typos and docs) Yes
• Did you verify new and existing tests pass locally with your changes? Yes
• Did you list all the breaking changes introduced by this pull request? None
• Did you update the CHANGELOG? (not for typos, docs, test updates, or minor internal changes/refactors) Yes

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

---

📚 Documentation preview 📚: https://pytorch-lightning--21307.org.readthedocs.build/en/21307/

[SkafteNicki]

Could we please also add some testing that the changed logic then correctly updates the learning rate when response=-1?

[LTMeyer]

Could we please also add some testing that the changed logic then correctly updates the learning rate when response=-1?

@SkafteNicki I've just added tests. Not sure about the most appropriate location though.
10d18d9 actually added tests:

• to check last_batch is set when rest of epoch was skipped.
• to check lr is being updated at the end of epoch when on_train_batch_start returns -1.

I've checked that these tests were indeed failing before the changes introduced in the current PR.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 79%. Comparing base (d046a96) to head (e3c416c).
✅ All tests successful. No failed tests found.

❗ There is a different number of reports uploaded between BASE (d046a96) and HEAD (e3c416c). Click for more details.

HEAD has 108 uploads less than BASE

Flag	BASE (d046a96)	HEAD (e3c416c)
cpu	54	27
python	6	3
lightning_fabric	12	0
pytest	27	0
python3.12.7	18	9
lightning	30	15
python3.11	12	6
python3.10	12	6
python3.12	6	3

Additional details and impacted files

@@ Coverage Diff @@ ## master #21307 +/- ## ========================================= - Coverage 87% 79% -8%  ========================================= Files 269 266 -3 Lines 23744 23692 -52 ========================================= - Hits 20569 18675 -1894  - Misses 3175 5017 +1842