Add more fine-grained status messages

[ospillinger]

Statuses

• live (including updating due to autoscaling) (for now just when min count is updated and ready)
• updating
• error
• compute unavailable
• out of memory
• stopping (bring it back)
• stopped

error -> updating -> live
live -> updating -> error

Guiding principles

• treat a deployment as a lambda with regards to e.g. autoscaling
• "status" is the answer to "what was the result of my deploy?"

Motivation

Currently, kubectl provides a better picture of what the cluster is working on than the CLI. Some deployments take a long time due to many python packages / large model / autoscaling and the user should be informed.