How to scale message processing on the Kafka consumer side; what is a partition and how are messages distributed across partitions?