How "low-level" management of threads is justified in Mirror Maker?

[Kostya Golikov]

I've skimmed through a source code of Mirror Maker and I see that it relies
on raw threads
<https://github.com/apache/kafka/blob/7df39e0394a6fd2f26e8f12768ebf7fecd56e3da/core/src/main/scala/kafka/tools/MirrorMaker.scala#L285-L287>,
as opposed to "Runnable + Thread pool" combination. While former approach
is just as functional as later, I am quite curious why such decision was
done. Anything specific that prevents move to runnables?