"solve_inG" is very slow for larger RA combinators

In his Iris fork, @jtassaro has a rather large RA that completely explodes solve_inG. Before !498 (merged), it took around 15s, and then Qed took another 15s.

Since !498 (merged), they just don't terminate any more. I can make the proof itself go through by adding a simpl in * in solve_inG (before this intros), but then Qed still takes forever. To make Qed work I added some awful hack that seemingly successfully guides Coq towards using the right reduction strategy for Qed.

I am not sure what we can do here, besides entirely sealing functors and thus making them non-computational (so we'd have to prove unfolding lemmas about them all, and somehow use them in solve_inG).