MultiIndex with ordered Categorical level should (maybe) respect ordered

[TomAugspurger]

Currently we lexsort. We should maybe follow the Categorical.ordered attribute. This is potentially a can of worms though

In [30]: import pandas as pd
    ...:
    ...: x = pd.Categorical(['apples', 'dairy', 'chicken', 'beef', 'apples', 'dairy', 'chicken'], categories=['apples', 'dairy', 'beef', 'chicken'], ordered=True)
    ...: y = pd.Series([1, 2, 1, 2, 1, 2, 1])
    ...: z = pd.Series([3, 4, 2, 1, 3, 2, 1])
    ...:
    ...: df = pd.DataFrame({'z': z, 'x': x, 'y':y})
    ...: df.set_index(['x', 'y']).sort_index()

Out[30]:
           z
x       y
apples  1  3
        1  3
beef    2  1
chicken 1  2
        1  1
dairy   2  4
        2  2

The index is lex-sorted. If it follows the categorical ordering, it'd be

           z
x       y
apples  1  3
        1  3
dairy   2  4
        2  2
beef    2  1
chicken 1  2
        1  1

cc @ thequackdaddy

[TomAugspurger]

From a consistency standpoint, we do follow Categorical.ordered when the index is just a single level, so we should probably follow it for MultiIndex too.

[TomAugspurger]

Closing in favor of #15058 (Joris had some additional points there)