Refactor Towards Log Subscription Perfection

[nzoschke]

This is available for testing with convox rack update 20160420210148-agent-0.68

Paired up with @awsmsrc to lick the problem where log the subscribers failed and never recovered, taking individual process logs offline until a restart.

• Refactor/rewrite subscribeLogs with much better Go concurrency and synchronization patterns
• Log with metrics around lines read from Docker (metric: Lines)
• Instrument, and log with metrics around events written to CloudWatch (metric: CloudWatchEventsSuccesses / CloudWatchEventsErrors)
• Normalize logging everywhere and add logging to to make it easier to understand in development
• Send all agent logs to the Rack LogGroup to better understand / debug and visualize in production
• Stop sending agent logs to the Rack Kinesis

This is a release candidate, though we could consider dialing down logs a bit before releasing this. I'll make a judgement call after some production instance hours.