Limiting the number of in-flight replacements across clusters

[brownleej]

We have operations that require a migration, such as changing the spec for the volume claims. These operations are going to require a temporary spike in resource usage between the time when we add new pods and remove the old pods, which can take a while. If someone updates the specs for multiple clusters in a way that triggers this change, they could exhaust their resource quota or their physical resources, putting clusters in a bad state. It might be useful to have an option to limit the number of in-flight replacements across clusters, to avoid bursting the overall resource usage too high compared to the steady state.