Cluster-ready

[smashwilson]

Running on a single host is dandy, but what we really want is to be able to spin up a cluster of these suckers, preferably adding and removing machines on the fly.

You can already run either the API server or the job runner instead of both by setting the PIPE_WEB or PIPE_RUNNER environment variables, and control where the job containers are actually launched with PIPE_DOCKERHOST and the TLS controls, so we're basically already there. Setting up the infrastructure to do this will be nontrivial, though.

[rgbkrk]

This may be a good chance to take docker/swarm for a ride.