gtid=oldest in kafkacdc is prone to breakage during binlog rotation

If the gtid=oldest mode is used in kafkacdc, it's possible that the GTID that was read is purged before the connection creation is successful. Since the binlogs are automatically rotated whenever the size exceeds the configured limit, this may happen more often than expected. An improvement to the way the GTIDs are handled would be to re-read the GTID after each failed connection attempt. This would make it so that a rotation that coincides with a connection attempt in kafkacdc would not cause it to stop.