Today I tried to replace a 200gb gist index with 8mb brin index and queries didn't work as expected - it was never used. set enable_seqscan=off helped for a bit but that's not a permanent solution.
A reasonable course of action on Postgres side seems to be to not assume selectivity of 1 in absence of correlation statistics, but something that would prefer such an index to a parallel seq scan, but higher than similar GIST.