BUG: .asfreq on resample on PeriodIndex/TimedeltaIndex are not

[jreback]

closes #12926

[jreback]

a tad hacky on this soln, as seems DatetimeIndex does lots of manipulations inside the binner to fix fencepost issues, while the other indexes (Period/Timedelta) don't seem to need this.

cc @MaximilianR created a structure to support common testing for resampling (the Base test class)

[max-sixty]

FYI this could super up

[jreback]

doesn't work. I think there is something going on w.r.t. how the values come back for DTI vs. TDI, meaning TDI always adds a stamp on (then removes it here). There is lots of internal logic, e.g. _adjust_bin_edges that is doing things.

[max-sixty]

Nice re the tests.

Tangential: I had a go deprecating the how keyword to make the Period stuff much simpler. I ran into issues with simple upsampling into super periods; I don't think that is possible without DatetimeIndex machinery atm. It's on my list though.

[jreback]

yep - I think we can still use much of the DatetimeIndexResampler machinery - just return the PeriodIndex at the end (and of course handle special cases)