Issue https://github.com/scylladb/scylla-enterprise/issues/5383 shows that we need to make CDC reader to slow down when it sees errors that signals about cluster/ node being overloaded.
So, we need backoff (exponential and flat) when driver replies with these errors:
BusyPoolException
OverloadedException
ReadTimeoutException
UPDATE
ExponentialRetryBackoffWithJitter
is already there, we need to expose it on configuration level and fix it to have proper backoff with small jitter.