Don't do state cleanup if training not converged and return trainer's state

[artemsok]

By default the training_state folder is deleted even if the training did not converge, i.e. the number of checkpoints did not hit --max-num-not-improved, but the training stopped because of reaching --max-samples, --max-num-epochs, or --max-updates. This changes the behaviour to keeping the folder to allow later training continuation and not enforcing the above parameters to match with the old run. Additionally a full training state is returned (with metrics, number of epochs, convergence status etc.), allowing its processing in a calling code.

Pull Request Checklist

• Changes are complete (if posting work-in-progress code, prefix your pull request title with '[WIP]'
  until you can check this box.
• Unit tests pass (pytest)
• System tests pass (pytest test/system)
• Passed code style checking (./style-check.sh)
• You have considered writing a test
• Updated major/minor version in sockeye/__init__.py. Major version bump if this is a backwards incompatible change.
• Updated CHANGELOG.md

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.

[fhieber]

Thanks for the change, makes a lot of sense to have this!

[fhieber]

LGTM