Fix: crmd: Don't update cluster/peer CIB fields when processing stonith results

Previously, to avoid tengine_stonith_callback() overwriting the CIB update
made by peer_update_callback() when a fenced node rejoins the cluster before
stonithd reports completion, the callback would set the sent_update flag.
However, tengine_stonith_notify() does not have access to this flag and
would update the CIB anyway.

Now, stonith_send_update() (used by both tengine_stonith_callback() and
tengine_stonith_notify()) doesn't set the cluster and peer fields in the
CIB at all (relying on membership layer and peer status callbacks to do so).

This way, the peer_status_callback()'s CIB update really doesn't get
overwritten, and peer_status_callback() doesn't have to try to do some
of the stonith callback's work (which would be erroneous if the node
was flapping and happened to rejoin before fencing actually occurred).