`deploy-all` should only actually deploy if all sub deployments have succeeded

[kammce]

This would prevent situations where users can break because we are currently deploying for all platforms and some finish before others. If one stalls for a while, the servers will have a recipe for that version but a missing prebuilt, giving the user a real prebuilt missing error. If we had a barrier for all of them or a way to defer the action to a follow up job with the same information, then we could make the deploy time very tight, reducing disruptions to developers.