We horizontally scale on queue depth/length (Kafka offset length specifically)

This requires the use of things like kubernetes custom metrics api and Prometheus adapter