Come up with a better way to handle interrupted connections

[Cryptophobia]

From @helgi on September 15, 2016 22:52

Most of the time Deployments continue when HTTP connection is severed between our CLI and our API. However, if bad config/image etc is pushed then we are not in as nice of a spot. Things like DB records and so on are not cleaned up and you end up in with an "In Progress" Deployment (see below for one possible solution)

Deployments have a few things on the Kubernetes roadmap (https://github.com/kubernetes/community/wiki/Roadmap:-Deployment) but there is nothing we can depend on in the near term

See deis/controller#1071 for more context

Copied from original issue: deis/controller#1074