Publishing a reply to a comment and then following it up with a second

reply would result in the reply banner sticking around, acting as if the

publish hadn't gone through.
This happened because signal handlers for the "publishing" and

"destroyed" events were registered twice on the first ReviewReply, which

would cause the state to mess up. We only really needed to register

these once.

Posted a reply, published. Posted another, published. Both times, the banner

disappeared.
Also tested with discarding.
Unit tests pass.