Arvados

While researching a customer question, I noticed that we have a config setting Containers.MaxComputeVMs which is not respected by arvados-dispatch-cloud. (It appears that it was previously used by the legacy "nodes" table and NodeManager). We should rename it to MaxInstances in the CloudVMs section and support it so users have a way to avoid blowing up their Arvados cluster by accident when they queue a really large workflow.