Add support for non-numeric types in Series.diff and DataFrame.diff

[Matt711]

This PR supports non-numeric data types (timestamp and ranges) in Series.diff and DataFrame.diff. In DataFrame.diff, datetime ranges are already supported because DataFrame.shift works. But Series.diff doesn't use the Series.shift implementation, so there wasn't support for datetime ranges.

import datetime
dti = pd.to_datetime(
    ["1/1/2018", np.datetime64("2018-01-01"), datetime.datetime(2018, 1, 1), datetime.datetime(2020, 1, 1)]
)
df = DataFrame({"dates": dti})
df.diff(periods=periods, axis=axis)

closes #10212.

[GPUtester]

Can one of the admins verify this patch?

[Matt711]

@bdice
For now, I replaced the way Series.diff is currently implemented (cudautils.gpu_diff.for_all) with Series.shift but that's not right. How should non-numeric types be supported in Series.diff?

[quasiben]

add to allowlist