[grouping] Precedence may be wrong when thread has stacktrace but exception doesn't

A customer is ending up with things being overgrouped because we're grouping on a generic error message from `exception` rather than the stacktrace in `threads`.  (Note that there are some questionable in-app rules being applied there - every one of those is a known system frame - but for the purposes of this issue all that matters is there are frames which we would use if we weren't ignoring the thread outright.) We should see about marking the exception component as non-contributing in this situation. (We'll need to figure out if that's always the right thing first.)