fix bad references

[emileten]

• partially closes KACE-1-0-G-tasmax.yaml fails clean-cmip6 with AttributeError: 'numpy.float64' object has no attribute 'year' from standardizing step downscaleCMIP6#439
• tests added / passed
• docs reflect changes
• entry in CHANGELOG.md

This fixes a few object reference mistakes in dodola.core.standardize_gcm, that were causing ClimateImpactLab/downscaleCMIP6#439 but could have been in general more problematic.

In the beginning of the function, a new transformed dataset object named ds_cleaned is created out of ds, but in subsequent calls, there are remaining references to ds where we should reference the transformed object instead.

This fix mechanically solves /ClimateImpactLab/downscaleCMIP6/issues/439 (which I verified in this workflow : https://argo.cildc6.org/archived-workflows/default/6da813ba-47ba-45a0-95d8-36444e2bcf7b), because we actually drop the time bounds in ds_cleaned, which were causing the issue.

[brews]

Nice fix, @emileten! This looks good to go. 👍

[brews]

I'm a little suprised that unit testing did catch this. Do we actually have a test checking that calendars get converted in standardize_gcm? This might be something to come back to later, as it sounds like an important behavior.

[emileten]

@brews, you mean that you're surprised it did not catch it, right ?

I don't fully understand your comment : we do have unit tests for calendar conversion. The 'error' I am fixing here wasn't as consequential as you might think, the calendar was still getting converted. Bad referencing meant that one step at the beginning of the code (where, among other things, we drop time_bnds) was not being taken in account. Since that step isn't 'key', things were in general working.

[brews]

@emileten Yeah, I was expecting a larger impact from the bug but I guess you're right. Thanks for the clarification.