This looks very interesting. I had real difficulty where I needed both a btree and gin(pg_trgm) index on the same column. When using `like` postgres would consistently choose the btree index which resulted in performance that was something like 15secs as opposed to the 200ms or so I'd see if the gin index were used. In the end I added two separate columns, one for each index so that I could force the correct one to be used for a particular query.