cftime: Fix resampling when time contains 0001-01-01

[dcherian]

• Closes cftime resampling error #9108
• Tests added
• User visible changes (including notable bug fixes) are documented in whats-new.rst

cc @spencerkclark

[spencerkclark]

Looks like CI confirms both of our experiences :).

I'm not sure why this would be platform-dependent...in the failing case I would be curious what ends up being returned from as_compatible_data in the traceback, and what step in as_compatible_data corrupts it. Unfortunately I don't think I have easy access to one of those platforms to do the debugging myself.

[dcherian]

Thanks @spencerkclark this has been quite useful. I can dig in a bit more on my Mac.

[spencerkclark]

Thanks @dcherian—I'll be curious what you find.

[dcherian]

Thanks @spencerkclark . Looks good to me!

[spencerkclark]

Thanks @dcherian for taking a look and starting the what's new entry! I made a slight wording change, since I think the test_cftime_resample_gh_9108 case might work with the old code if the resampling frequency were "MS" instead of "ME"—maybe you can confirm on your machine.

It was good to be able to address this and remove some old cruft, e.g. the _days_in_month and _shift_month functions date all the way back to #2301, which was prior to the daysinmonth property being implemented on cftime objects (Unidata/cftime#138) and the addition of the has_year_zero option (Unidata/cftime#234).

I'll go ahead and merge at the end of the day tomorrow if there are no objections.

[spencerkclark]

Thanks again for your help on this @dcherian