Overhaul moving distribution APIs

Mostly this means dropping the up call and just letting users use cdf(0) if they want to force cleaning a dirty CDF.

It also means bifurcating flat time kernel histograms into Hist and CHist where the latter cumulative variant does the least total work in update-every-point workloads, but the former does less in many updates then query workloads. As Mr. T might say, pity the fool who doesn't know which workload they have. ;-) (Well, really then, maybe a Bist is more balanced.)