refactor: modifying _update_from_checkpoint signature

[jeandut]

Related issue

No related issue

Summary

Notes

This PR introduces a change in the signature of the algorithm's update_from_checkpoint function in order to ease its use in complex pipelines.
As of before this PR if you had a checkpoint (a dict from say a call to get_state_to_save) it was impossible to load it ...
Now workflows can use patterns like:

chkpt = self.get_state_to_save()
self._update_from_checkpoint(chkpt)

Please check if the PR fulfills these requirements

• If the feature has an impact on the user experience, the changelog has been updated
• Tests for the changes have been added (for bug fixes / features)
• Docs have been added / updated (for bug fixes / features)
• The commit message follows the conventional commit specification

[SdgJlbl]

Thanks a lot for your PR.
I understand the context of the change, and I agree that it is best to separate the I/O (loading the checkpoint) and the update part. Nevertheless, the method _update_from_checkpoint was idempotent (i.e. you can apply it several times in a row and get the same results as if you apply it once), and it's not the case anymore after your change, because of the side effect on checkpoint. It would be safer to keep _update_from_checkpoint idempotent, either by reading from checkpoint but not popping values from it, or by passing a copy to the function.

[jeandut]

Okay will copy checkpoint in the function however striving for functional purity is a bit of an unattainable ideal.