When a database connection is lost while a connector is running, the exception will be thrown and retried due to how the BaseSourceTask#poll method works.

The problem is if the database connection isn't restored by the time the Kafka Connect environment restarts the connector and calls BaseSourceTask#start, the connector will enter a FAILED status and will be terminated because Kafka Connect does not support the notion of RetriableException handling during the start-up of a connector.

We should introduce a connection retry mechanism, similar to slot.max.retries and slot.retry.delay.ms but where we will attempt to retry connecting to the database from within the BaseSourceTask#start method before throwing an exception, making the lifecycle of the connector more resilient.

Settings these values to 0 would retain current behavior, and perhaps that would be the ideal default value during 2.0, but giving users the ability to opt-in to this retry behavior as needed?