Highly requested: cooperative graph draining with RunControl.
Request a drain from the caller or inside a node, let the current superstep finish, checkpoint the graph, then resume later from the same config.
Great for deploys, worker shutdowns, and long-running agent jobs.