fix resuming from checkpoint for fault-tolerant in case of no failure

[awaelchli]

What does this PR do?

Fixes #8835

When saving the loop state dict on a failure (breaking in the middle of an epoch), we want to save the progress counters exactly as they are in order to resume on the failed batch. The same applies to runs that finish successfully and with the full epoch completed. If we resume training in such a case, we don't want to resume on the last batch that we saved, but instead on the next one which is the batch 0 from the next epoch. So the counters need to reset every start of an epoch if:

• we are not restarting (regular case)
• if we are restarting, the current counters are at the limit of the loop AND we did not fail

Note:

• This fix applies to master branch only (fault-tolerant training).
• Test fails on master

TODO:

• test

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 🙃

[tchaton]

LGTM !

[justusschock]

LGTM!

[codecov]

Codecov Report

Merging #9371 (8045fd5) into master (81687aa) will decrease coverage by 0%.
The diff coverage is 100%.

@@          Coverage Diff           @@
##           master   #9371   +/-   ##
======================================
- Coverage      93%     93%   -0%     
======================================
  Files         180     180           
  Lines       14965   15046   +81     
======================================
+ Hits        13904   13971   +67     
- Misses       1061    1075   +14