[FEA] Series datetime ceil

[beckernick]

As a complement to Series.dt.floor (#7102), I'd like to be able to use call ceil to round the frequency level of the data to a specified frequency (noting that ceil implies "rounding up" to the desired frequency, while floor implies rounding down).

Based on comments in #7102 , this might require libcudf support.

import pandas as pd
import numpy as np
​
np.random.seed(12)
​
s = pd.Series(pd.date_range('1/1/2018 11:59:00', periods=1000, freq='s')).sample(3)
s
518   2018-01-01 12:07:38
871   2018-01-01 12:13:31
797   2018-01-01 12:12:17
dtype: datetime64[ns]

s.dt.ceil('H') # nearest hour (round up)
518   2018-01-01 13:00:00
871   2018-01-01 13:00:00
797   2018-01-01 13:00:00
dtype: datetime64[ns]

s.dt.ceil('D') # nearest day (round up)
518   2018-01-02
871   2018-01-02
797   2018-01-02
dtype: datetime64[ns]

s.dt.ceil('min') # nearest minute (round up)
518   2018-01-01 12:08:00
871   2018-01-01 12:14:00
797   2018-01-01 12:13:00
dtype: datetime64[ns]

[beckernick]

With #8942 merged, the next step is Cython bindings for the Python layer. @shaneding should someone else pick that up, as I believe you're doing some Spark plugin work currently?