Make Master more robust with state persistence

[nilesh-c]

If the Master dies and restarts, we lose state.

Akka supports something called PersistentActor that lets us save actor state. Worker-side state is strictly unimportant. If a Worker dies and restarts, the corresponding job is just re-enqueued by the Master and everything goes on as usual.

If a Master fails we have a problem. We loose the queue, mirrors statuses, worker progress reports, and worker download size (bytes) distribution. Using PersistentActor can help us handle this easily. It uses an append-only "journal" to record state (events rather!) to.

I am yet to come across a situation where my Master actor died, but this is certainly a possibility that we should handle. Any loose Exception can kill the Master.