Minor Description
The `Worker` class has an `executor` field that the public constructor initializes with a new cached thread pool (https://github.com/apache/kafka/blob/02226fa090513882b9229ac834fd493d71ae6d96/connect/runtime/src/main/java/org/apache/kafka/connect/runtime/Worker.java#L127.]).
When the worker is stopped, it does not shutdown this executor. This is normally okay in the Connect runtime and MirrorMaker 2 runtimes, because the worker is stopped only when the JVM is stopped (via the shutdown hook in the herders).
However, we instantiate and stop the herder many times in our integration tests, and this means we're not necessarily shutting down the herder's executor. Normally this won't hurt, as long as all of the runnables that the executor threads run actually do terminate. But it's possible those threads might not terminate in all tests. TBH, I don't know that such cases actually exist.