I played a bit this week, and at least the crash for the score in
https://lists.gnu.org/archive/html/lilypond-user/2022-11/msg00272.html
(which is finally a crash that I can reproduce) seems to be solved by
the current tip of the release-8_2 branch. A minimal backport of that
commit on its own also seems to work. However, my original plan was to
keep the current workaround in place *and* backport a minimal version
of the fix. For some inexplicable reasons, this avoids the crash but
then just hangs after printing the warning (that I did not patch). This
doesn't make me feel comfortable doing any last minute change in that
area, potentially breaking things even worse, and I will not do it for
this weekend.