Fix for matches being lost near the celestial poles

[stargaser]

This PR fixes the drop in crossmatching density near the celestial poles by applying the cosine of Declination following astronomy-commons#22 and astronomy-commons#23.

The fix for overflowing histogram2d from astronomy-commons@56248a1 has been included here as I wanted to test a subset of the sky near the celestial poles.

[stargaser]

I have come to the realization that the Sort-Merge-Join speedup requires a single table column to appear in the range expression for matching. I've added an racosdec column that is the product of right ascension and the cosine of declination. This clears up the hole in the celestial poles! However, matching across the RA 0 to 360 boundary is even more of a problem than in the current implementation.

I think all problems can be solved by adding the racosdec column, in addition to zone and dup, and by duplicating RA values within NGBR_BORDER_HEIGHT (10 arcsec) of 0 deg and 360 deg across the boundary. This does have the drawback of storing another column with every table. I don't see an alternative way out, but perhaps @ctslater has ideas?