complement for issue 2097 : total update fails sending a Message to ReplicationCache

When debugging 2097 I've noticed that during a replication total update,
if the updated server fails, the publisher thread on the other server hangs.

The hang happens because normally the publisher thread never tries to reconnect
after a protocol session failure but waits for the receiver threads to do the job but
here the publishing activity is done in a receiver thread that holding the session lock
and therefore prevent the other thread from doing there job.

The solution is to move this work outside of the lock.

I've tested this manually as it is difficult to automate.