Implement sort_remaining for sort_index

[wence-]

Description

Previously, the sort_remaining argument to sort_index was ignored. Passing sort_remaining=False would raise a NotImplementedError. Moreover, for a multiindex, sort_remaining=True was not handled correctly: if not all levels were requested as sorted, sort_index would behave as if sort_remaining=False had been passed.

To fix this case, construct the sort order based on first the provided levels and, if sort_remaining=True, the left-over levels (in index order).

To facilitate this, refactor the internal _get_columns_by_label function to always return a Frame-like object (previously, if we had a Frame we would get back a ColumnAccessor, and it was only for IndexedFrame and above that we'd get something of Self-like type back). This meant that calling _get_sorted_inds with by != None was not possible on an Index or MultiIndex (the code assumed we'd get a Frame back).

• Closes [FEA] Allow DataFrame.sort_index(level=n) to tiebreak on lexsortedness in mode.pandas_compatable #14011

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[galipremsagar]

LGTM, minor comment.

[wence-]

/merge