How would you speed up regression from a set of Kafka scenarios where waiting for processing is tied to timeouts?