Fixes #11099 : Indexes lost upon .swap_dims() call

[Reshief]

Indexes of non-dimension variable names being lost upon ds.swap_dims() call

This is a small patch to fix the issue underlying #11099 .
Due to Indexes only being copies in Dataset.swap_dims() for variables whose names are dimensions before the call, indexes for multi-index variables and probably also for other indexed variables are lost in the resulting dataset.
This can lead to issues further down the line like .attrs being lost (as I personally encountered).

The patch copies over index variables and indexes of variables that do not lose dimension-status, making the function .swap_dims() behave consistently with .rename_dims() if the target name is not used.

I added a short test to make sure that the patch fixes the reported problem in #11099 and ran the entire test suite to make sure that no unintended changes to overall behavior that has previously been tested would be observed.

• Closes .swap_dims() loses indexes of non-dimension variables #11099
• Tests added

[dcherian]

Thanks, seems OK to me.

If you want fix some related bugs there are a few documented here:

xarray/properties/test_index_manipulation.py

Lines 244 to 249 in 49ec516

	# TODO: swap_dims is basically all broken if a multiindex is present
	# TODO: Avoid swapping from Index to a MultiIndex level
	# TODO: Avoid swapping from MultiIndex to a level of the same MultiIndex
	# TODO: Avoid swapping when a MultiIndex is present
	@precondition(lambda self: not bool(self.multi_indexed_dims))
	@precondition(lambda self: bool(self.swappable_dims))

Removing @precondition(lambda self: not bool(self.multi_indexed_dims)) should reproduce them for you