So far dma would sleep for the whole backoff period before retrying a
delivery. However, this also meant that a user issuing `dma -q` could
not force these mails to be retried immediately.
Switch to sleeping for only 30 seconds at a time, and poll the mtime
of a special queue flush signal file `flush` in the spool directory.
If the flush file has been touched since the beginning of the backoff
period, initiate a retry right away.
When flushing the queue with `dma -q` we now touch the flush file to
make waiting instances of dma process this user request.
The 30 second poll interval at the same time avoids a DoS via
excessive `dma -q` by any user.