Fix Dataset.map to preserve attrs set by the applied function

[C1-BA-B1-F3]

Description

This PR fixes #11356 by allowing Dataset.map to preserve attributes that are explicitly set by the applied function.

Problem

Previously, when a function passed to Dataset.map explicitly set attributes on a DataArray (e.g., via .assign_attrs()), those attrs were lost because:

• With keep_attrs=True: original attrs were copied back, overwriting func's changes
• With keep_attrs=False: all attrs were wiped to empty dicts

This made it impossible to use Dataset.map to update variable attributes.

Solution

Now, if the function returns a DataArray with non-empty attrs that differ from the original, those attrs are preserved regardless of the keep_attrs setting. This allows users to update variable attributes through Dataset.map:

ds = xr.Dataset({
    'a': ('x', [1, 2, 3], {'units': 'kg'}),
    'b': ('x', [4, 5, 6], {'units': 'kg'}),
})

# Now works correctly - attrs are preserved
result = ds.map(lambda x: (x / x.sum()).assign_attrs(units='unitless'), keep_attrs=True)
assert result['a'].attrs == {'units': 'unitless'}

The existing behavior is preserved for all other cases:

• keep_attrs=True still restores original attrs when func doesn't change them
• keep_attrs=False still wipes attrs when func doesn't change them
• Functions that explicitly drop attrs (via .drop_attrs()) still follow the keep_attrs logic

Tests Added

• Regression test test_map_preserves_func_attrs covering all four scenarios:
  1. keep_attrs=True with func that sets attrs → func's attrs preserved
  2. keep_attrs=False with func that sets attrs → func's attrs preserved
  3. keep_attrs=True with func that doesn't set attrs → original attrs restored
  4. keep_attrs=False with func that doesn't set attrs → attrs wiped

Fixes #11356

[welcome]

Thank you for opening this pull request! It may take us a few days to respond here, so thank you for being patient.
If you have questions, some answers may be found in our contributing guidelines.