Allow to issue a cancel/kill command and retry immediately after the kill/cancel happened. This can be useful in many situations, e.g.:

• You want to retain invocation ordering on keyed services, but one invocation is stuck. You fix the service, deploy it, and then execute cancel/kill and retry (you might have to be able to delay the execute the old cancellations code).
• You need to retry, but the invocation came from another service and you don't have the payload of the request at hand for retrying it manually