Try random page cost 1.1. Way back when I started using SSD we had a
discussion here and came to the conclusion that it should be ever so
slightly higher than sequential page cost.
It is very hard to read your query plans (maybe gmail is wrapping them
funny or you need to use a fixed font on them or share them from
https://explain.depesz.com), but they do look substantially different. My
guess is that with the random page cost = sequential page cost you are
tricking Pg into using more sequential scans than index searches.