The goal of this is to centralize the polling logic in a single class and also separate the queue writers (e.g. binlog reader) from the queue readers (the code polling from the queue and propagating the changes to the Kafka Connect framework. It will also allow us to implement handling of certain options in a single place.

This will also allow us to get rid of the BlockingReader in the MySQL connector: if the snapshot reader is done, the polling loop will keep going, but the handling in ChangeEventQueue#poll() will avoid the high CPU load originally motivating the blocking reader, as it regular sleeps for a while.