of local sockets and pipes all open at once, and if one of them is
uncleanly closed from the remote end we don't want the whole
application to die - we want to close that socket's SSH channel and
continue with the rest of the run.
[originally from svn r9359]