Kafka middleware: concurrency control

Add configurable parallelism — e.g. process N messages concurrently per partition, or across partitions. Currently messages are processed one at a time, sequentially per partition.

Related: #177 (batch processing — both address throughput, design choices overlap).

Start at the Kafka middleware entry point and trace how messages are currently processed sequentially within partitions. Read related issue #177 to compare the overlapping throughput and design concerns. Done means a documented, configurable concurrency model that processes messages in parallel while preserving the required partition behavior.

Written by the indexing model from the issue text.