Improve speed such that a Babel diff run takes less than 12h

[gaurav]

There is a 12h runtime limit on our cluster. This was just good enough for the previous versions of Babel, but we go over this limit with the larger Babel that now includes the UMLS compendium. We can tweak our way around this (e.g. by creating multiple batches that process different subsets of the compendia), but of course it would be great if we could speed this up.