Factor out the debug logger

[kriskowal]

Replace all log messages with events and echo all worker events on the cluster supervisor itself. Create a tool that channels events to a logger to the same effect as we have now.