Moving a busy account means moving it twice: the bulk ahead of time, then
a short catch-up at the cutover so almost no mail is lost to downtime. The
import and export steps already converge -- each run moves only what
changed -- so the catch-up was there, but driving it was manual: run,
wait, run again.
--repeat <DURATION> drives it. After each pass over the accounts it waits
that long and runs them again, each pass moving only what has arrived at
the source since the last, until the run is interrupted. The setup runs
once, before the first pass. Leave it running while the old server still
takes mail; change the MX and stop it, and the final pass is a short
catch-up, not the whole mailbox. A pass in flight is resumable, so Ctrl-C
is safe.
--max-passes <N> bounds it, and --until-quiet stops on the first pass that
moves nothing new -- the source has gone quiet and the servers are in
step, a cue the cutover is safe. Each pass re-runs every step, like
--again, which is what makes it catch the delta rather than skip the
done steps.
The account pass is factored into run_accounts_pass and the run loops it;
Outcome gains the pass count, the json progress gains a per-pass end event
and a waiting event between passes. Tests cover the multi-pass loop, the
--until-quiet stop and the duration parser.