Time based validation support

[Sohaib-Ahmed21]

What does this PR do?

• Adds an option to run validation by time (e.g., every 15 seconds) via val_check_interval (string, timedelta, or dict).
• The timer starts at fit, triggers validation when elapsed, respects check_val_every_n_epoch, and resets after each validation.
• In time mode we skip epoch-end checkpoints unless save_on_train_epoch_end is set to True.
• Existing int/float behavior stays the same.
• Docs updated and tested by building them.
• Tests updated.

Fixes #13324

Before submitting

• Was this discussed/agreed via a GitHub issue? (not for typos and docs)
  yes
• 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)
  Yes
• 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?
• Did you list all the breaking changes introduced by this pull request?
  No breaking changes introduced
• 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:

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--21071.org.readthedocs.build/en/21071/

[Borda]

I like the extension and I think it valuable use-case for continues training, but leaving the decision to @lantiga

[lantiga]

Thank you for the contribution @Sohaib-Ahmed21

Goes without saying, runs become non reproducible, as in:

• it's not possible to re-run and have validation happen at the same intervals
• interrupting a run and resuming will not lead to the same validation curves

I would make sure users understand this, so in case they use validation loss as a criterion for scheduling or stopping they will be fully aware.