runFinalClose ran the transcript archive, creator DM, close log, and closure
email in the same try as the close transition and channel delete, with the
transcript posted *before* the commit. A failure in any of them (notably a
DiscordAPIError 50001 Missing Access when posting the transcript to the archive
channel) aborted the whole close: the customer saw 'ticket closed' but the DB
stayed open and the channel was never deleted.
Rewrite so the close transition + pendingDelete commit FIRST, each side-effect is
individually best-effort via a closeStep wrapper, and scheduleTicketChannelDelete
always runs. finalizeForceClose was already commit-first; wrap its remaining
unguarded archiving send too.