Skip inference on EFO

[ppavlidis]

In discussions with @neerapatadia it has become clearer that doing query expansion via EFO inference is a problem.

The issue is that EFO is extremely flat with many heterogeneous terms being siblings. (we should try to find alternatives terms in other ontologies if that is an issue) Our naive inference approach means we group contrasts together that are not thematically related.

Using a more motivated approach like information content to avoid propagating to low-information nodes might be good to consider as well, but a heuristic for EFO seems reasonable, compared to a stopword list.