There are a number of cases where queries in both pgadmin4 and pgadmin3 are done against pg_depend but falis to constrain the query on classid.
In particular, if for example a constraint and a sequence exists with the same oid (which is perfectly valid, as they are in different tables), a column will suddenly refer sequences that are completely incorrect. When we look up sequences, we have to make sure we only match dependencies against pg_class.
AFAICT, the following patch catches the important ones. I could go ahead and push it, but given that we're in RC I'd appreciate a round of review before doing that, even though it looks fairly simple.
I've got the same changes in a patch to pgadmin3, as these queries are basically a copy/paste from there. But let's focus the review on the one for pgadmin4 :)
--