Handle retry of event groups that span multiple relay log files.

 - If retry reaches the end of one relay log file, move on to the next.

 - Handle refcounting of relay log files, and avoid purging relay log
   files until all event groups have completed that might have needed
   them for transaction retry.