I have been running a cron script to automatically restart the lemmy backend, which in turn resets the postgres memory use ever since this problem started to happen months ago. For me 0.19.x actually made it less bad, but it is still an annoying issue.