Kill the container ourselves when done

[shepmaster]

We stream messages from the coordinator to the worker via stdin/stdout piped to docker run. However, sometimes we close the pipe corresponding to docker run's stdout before it has had a chance to fully write everything. If that happens, docker run exits without deleting the container 1.

This has likely always been a problem with the coordinator / worker architecture, but it's exacerbated by the recent addition of the memory / CPU status messages that occur every second, which greatly increase the amount of data crossing the boundary.

This reorganizes the code to pass in a known reasonably unique name to the container and then attempts to kill that container at cleanup time.

I tested by:

• Setting the worker status interval to 16ms (from 1s)
• Setting the webserver idle interval to 7s (from 60s)
• Setting the webserver session timeout to 14s (from 45m)
• Running code that prints infinitely (loop { println!("moo"); })

This reliably recreated the write /dev/stdout: broken pipe error from the docker run invocation, but after this commit there were no vestigial running containers.