ENH: cut return bins should be Timestamp/Timedelta objects for datelike data

[mroeschke]

In [1]: s = pd.Series(pd.date_range('20180101', periods=3))

In [2]: s
Out[2]:
0   2018-01-01
1   2018-01-02
2   2018-01-03
dtype: datetime64[ns]

In [3]: res, bins = pd.cut(s, 2, retbins=True)

In [4]: pd.cut(s, bins)  # Bins do not "roundtrip"

ValueError: bins must be of datetime64 dtype

In [5]: pd.cut(s, [pd.Timestamp(b) for b in bins]) == res
Out[5]:
0    True
1    True
2    True
dtype: bool

For convenience and argument of "roundtripping", I think cut(..., retbins=True) with datelike data (datetime or timedelta data) should return bins that is an array of Timestamp/Timedelta objects instead of an array of integers.

[jreback]

related to #19872