API: Timestamp/Timedelta resolution in __repr__

[jbrockmendel]

Discussed on today's call, in 2.0 Timestamp and Timedelta will support non-nanosecond resolutions, and this should probably be reflected in the repr, something like Timestamp("2016-01-01 10:24:11.012", reso="ms").

ATM it is possible to do eval(repr(ts)) on a Timestamp. If we want to maintain that, we'll need to add a reso keyword (or another name, we discussed "unit" but that conflicts with the existing unit keyword) to the constructors.

If we do add a reso keyword, it may make sense to add it to to_datetime, date_range, to_timedelta, timedelta_range. That would avoid the need to do reso-inference in those functions.

[jreback]

adding reso to the constructors and repr lgtm