Make handleAnswerPrimaryMaster, handleRequestNodeIdentification and handleAskPrimaryMaster resilient to connection loss. As they are involved in question-response exchanges during election on connections which will get dropped as soon as a master arises, they must check if the connection is still opened before trying to send.
Note: the connection might get closed anyway during the actual send, but this is handled separately already. The problem we fix here emerged because of more tedious checkings in Connection's ask & answer methods.


git-svn-id: https://svn.erp5.org/repos/neo/branches/prototype3@799 71dcc9de-d417-0410-9af5-da40c76e7ee4