Master subprocess watcher.

Start and watch the configured runners and ensure that they stay alive and
kicking.  Each runner is forked and exec'd in turn, with the master waiting on
their process ids.  When it detects a child runner has exited, it may restart
it.

The runners respond to SIGINT, SIGTERM, SIGUSR1 and SIGHUP.  SIGINT, SIGTERM
and SIGUSR1 all cause a runner to exit cleanly.  The master will restart
runners that have exited due to a SIGUSR1 or some kind of other exit condition
(say because of an uncaught exception).  SIGHUP causes the master and the
runners to close their log files, and reopen then upon the next printed
message.

The master also responds to SIGINT, SIGTERM, SIGUSR1 and SIGHUP, which it
simply passes on to the runners.  Note that the master will close and reopen
its own log files on receipt of a SIGHUP.  The master also leaves its own
process id in the file `data/master.pid` but you normally don't need to use
this pid directly.

master: error: The master lock could not be acquired, because it appears as if some process
on some other host may have acquired it.  We can't test for stale locks across
host boundaries, so you'll have to clean this up manually.

Lock file: /home/cce/repo/mailman/var/locks/master.lck
Lock host: linux-35.local

Exiting.