Choose which dimensions are reduced in measures_improvement

[coxipi]

Pull Request Checklist:

• This PR addresses an already opened issue (for bug fixes / features)
  • This PR fixes #xyz
• (If applicable) Documentation has been added / updated (for bug fixes / features).
• (If applicable) Tests have been added.
• This PR does not seem to break the templates.
• CHANGELOG.rst has been updated (with summary of main changes).
  • Link to issue (:issue:number) and pull request (:pull:number) has been added.

What kind of change does this PR introduce?

• measures_improvement has an extra kwarg: dim, which allows to control on which dimensions we compute the percentage of improvement.

Does this PR introduce a breaking change?

Default behaviour None is the same as before, reduce all dimensions. I thought it made sense to put dim as the first kwarg, so this is not breaking per-se, but still somewhat semi-breaking for scripts inputting to_level positionnally e.g. measures_improvement([ds1,ds2], to_level), so I can move down dim if it's preferrable.

There is a small breaking change: ds1-ds2 is used instead of ds2 to find non-null values and compute the pct of improvement. This may result in less non-null values. I think it's the right thing to do though.

Other information:

My example is that I had a dataset with (rlat, rlon, period), I want improvement percentages for each period. I realize now this is an unconventional dataset in the xscen philosophy, I should have had separate dataset, when using properties_and_measure I should have used period etc etc. and have multiple catalog entries. But, for properties using group=time.season, time.month, we would have an extra dimension season, month that we don't necessarily want to reduce, so I think it still makes sense.

Also: I think this is unrelated to templates, I assumed this is related to catalogs? So I checked the box

[github-actions]

Welcome, new contributor!

It appears that this is your first Pull Request. To give credit where it's due, we ask that you add your information to the AUTHORS.rst and .zenodo.json:

• The relevant author information has been added to AUTHORS.rst and .zenodo.json.

Please make sure you've read our contributing guide. We look forward to reviewing your Pull Request shortly ✨

[coxipi]

For now, I assume that if dim is precised, it is present in every variable. If we want even more control, we could accept dim as a dict, where every key is a given variable. I think this would not be used that often. With dim=None, all dimensions are reduced like before, regardless if dimensions are different in distinct indicators (e.g. (time) vs. (time,season))

[RondeauG]

LGTM, but I'll let @juliettelavoie approve the PR since she's a lot more familiar with this part of the code.

[coxipi]

The current test fail is in TestAdjust::test_basic which should not be affected by my changes

[juliettelavoie]

super!
I think the breaking changes for the nan and positional to_level are minor enough.

FYI, The templates are here: https://github.com/Ouranosinc/xscen/tree/main/templates
the first ones uses measures_improvement, but your changes are not breaking enough to break that script.