fix the bank splitter bug that group random subbanks.

We recently discovered the current bank splitter accidentally bundle sub-banks coming from random range in \chi_{eff}. See the violin plot below (200 templates * 5 sub-banks).

I found that this line is re-sorting sub-banks based on the parameter set by options.sort_by regardless of \chi_{eff}. Rather, we want them to be sorted inside each \chi_{eff} bin.
So I made a change so that it groups multiple sub-banks into one svd bank in each \chi_{eff} and that svd bins are still sorted by options.sort_by afterwards, which gives us better sense of which svd id roughly corresponds to BNS/NSBH/BBH.
After this fix, the violin plot looks much better.