ENH: `RangeIndex.sort_values` can avoid materialization of values.

[galipremsagar]

Is your feature request related to a problem?

We can avoid the materialization of RangeIndex when sort_values is invoked on it. The reason being RangeIndex is always sorted.

Describe the solution you'd like

>>> import pandas as pd
>>> pd.__version__
'1.3.3'
>>> idx = pd.RangeIndex(0, 10, 1)
>>> idx
RangeIndex(start=0, stop=10, step=1)
>>> idx.sort_values(ascending=True)
Int64Index([0, 1, 2, 3, 4, 5, 6, 7, 8, 9], dtype='int64')  # Instead we should be returning RangeIndex(start=0, stop=10, step=1)

>>> idx.sort_values(ascending=False)
Int64Index([9, 8, 7, 6, 5, 4, 3, 2, 1, 0], dtype='int64')    #Instead we can return RangeIndex(start=10-1, start=0-1, step=-1), see below
>>> pd.RangeIndex(10-1, 0-1, -1)._values
array([9, 8, 7, 6, 5, 4, 3, 2, 1, 0])

API breaking implications

This would break the expected return types of Index of DataFrame & Series in some operations, but would result in speedup of operation as we would not be materializing the RangeIndex.

[galipremsagar]

Note we already have this kind of behaviour to premptively avoid materialization in DataFrame.sort_index:

>>> df = pd.DataFrame({'a':[1, 2, 3]})
>>> df
   a
0  1
1  2
2  3
>>> df.sort_index()
   a
0  1
1  2
2  3
>>> df.sort_index().index
RangeIndex(start=0, stop=3, step=1)

Hence requesting if we can make this behavior uniform across all APIs by incorporating it in Index.sort_values.

[usersblock]

take

[usersblock]

This should only make a new RangeIndex if it is being converted from ascending to descending and vice-versa.