propagate attrs on coords in Dataset.map

[keewis]

• Closes Missing coordinate attributes after weighted mean #9317, closes Fix missing coordinate attributes after apply_ufunc operations #10083
• Tests added
• User visible changes (including notable bug fixes) are documented in whats-new.rst

This makes sure we propagate attrs in coords in Dataset.map if keep_attrs == True. It uses the same semantics as for the data variables, which means that keep_attrs=True will overwrite any changes to the attrs by the mapped function.

This hides the use of the division operator in some of the weighted operations, which are solely affected by the global keep_attrs property.

Alternative implementation options:

• Instead of collecting and merging the coords then copying over coordinate attrs we could also make DataArray._copy_attrs_from iterate over the coordinates.
• Instead of modifying Dataset.map, we could also make all weighted ops receive a keep_attrs parameter

[dcherian]

Instead of modifying Dataset.map, we could also make all weighted ops receive a keep_attrs parameter

I think this is a good improvement for Dataset.map. I don't think weighted should take keep_attrs, we've tried to leave that control to the actual reduction (.weighted.mean) : #4513