(For decisive answers, read the postgres code ;))

And to be sure, please use like '%ab%' again.

I can't get bigrams to respond quickly; also not when there is only one matching value; with this data it will always Seq scan (sometimes with parallel workers: just under 1 second).

In postgres, 'LIKE' doesn't allow regex (although its simple pattern search can sometimes use the trigram or btree index). Postgres uses the tilde for regex search (~ case sensitive, ~* case insensitive).