We need to do some db cleanup before exiting workers (same issues seen #167).
The old implementation calls process.nextTick(process.exit()) - which prevents any new code on the event queue from running after the event is emitted, but before process.exit() is called.
Also, it appears as though the 'close' event was never propagated up to the master. I decided to propagate 'worker closed' instead.
This closes #167
* emit 'worker close' in master process when a worker closes
* delay execution of process.exit with a callback when listening on 'worker close'