There's a few things which should be improved after the initial work on this offset storage implementation (via DBZ-4509):

• Move the implementation into a separate module so it can be used solely with Debezium Engine, but without Debezium Server
• Investigate why the scope of the PG connector dependency had to be changed to "runtime" rather than "test" (see discussion in PR 3220)
• Move to contract independent of Kafka Connect once available (see DBZ-1971)