Improving (re-refinement) of bin-refine factors with high number density datasets

[manodeep]

Currently, when default bin refine factors are used, the code tries to increase the refinement in along one axis. A better strategy to be to recursively apply the refinement (or estimate the level of refinement required depending on the current refinement factors and the average number density).

[lgarrison]

I think we boost along two axes in most cases to help with this, e.g. https://github.com/manodeep/Corrfunc/blob/master/theory/DD/countpairs_impl.c.src#L252-L255

But I agree, we should just compute a target density and set the bin refine factors to match that.