Improve deployment stage

[michaelsauter]

There are a couple of improvements I'm planning to implement for the deployment stage:

• Right now, the deployment assumes that ImageStream and DeploymentConfig of a certain name exist, and it does not fail in a friendly way if this is not the case.
• The deployment happens because the latest tag is updated. The command used for that is deprecated (see Tagging images is deprecated #67), and more generally, it assumes that an image trigger is present which might not be the case.
• There is no deployment timeout - right now the step can take very long and the user cannot control how long it is allowed to take.
• Since 1.2, the result of the deployment is checked (which is awesome as it fails the build when the deployment fails - combined with health checks that allows for a very robust rollout), however the code relies on a lot of oc output, which is brittle (see e.g. Seldom error in stageDeployToOpenshift (ArrayIndexOutOfBounds)- when checking for new deployment #142) and does quite a few calls, which can be optimised.
• Finally, it would be nice to expose the "building blocks" of deployment so that quickstarters which need more control over the deployment (e.g. because they have two DeploymentConfig resources) can use those building blocks.

I have a prototype of this ready and will give it some test time this week and then open a PR.