fix: Make rochester_lookup compatible with dask_awkward

[jrueb]

Before, rochester_lookup.kSmearMC and rochester_lookup.kSmearMCerror would not work with dask-awkward arrays. This was because of a call to doublecrystalball.ppf (scipy) and numpy.zeros_like.
Fixes #871

[lgray]

@ssrothman FYI we got the old version working - we should settle on a plan to correctionlib-ify rochester corrections.

The tool in coffea to extract the current rochester corrections into something pythonic should be a good starting point for onboarding since you got a good bit of the rest of it done iiuc.

@nsmith- @andrzejnovak @NJManganelli - adding you here since we should keep in mind that we'd like to go this direction (correctionlib rochester)

[lgray]

Let me put the tests in dask and we'll be good to go I think.

[jrueb]

I fixed the test a bit. The issue was that the first argument to ak.unflatten was a numpy array, while the second one was a dask-awkward array. This is not possible with awkward yet.
However, the test now seems to fail because of dask-contrib/dask-awkward#313, which is still left unsolved.

[lgray]

Thanks I'll take a look. I had to pause on this to take care of other issues.

[lgray]

@jrueb just checking in here - the underlying issues in awkward / dask-awkward that are impeding progress here are being worked on my Angus and Jim. But it seems we wait on those for now.

[lgray]

So with some fixes from awkward and dask-awkward this fails in a different way. I'll keep using this to exercise the changes coming down the pipeline that aim to fix the present failure.

[lgray]

@jrueb FYI scikit-hep/awkward#2652 now fixes the failures in this PR.