Fix: controller: update node state correctly based on any existing node cache entry

Previously when controller was updating node state for a fencing result,
it'd search for any node cache entry purely by node name. But probably
an existing node cache entry didn't have an uname yet. Then
crm__get_peer_full() could not find it and would create yet another node
cache entry which didn't have the correct node state though.

In a bad situation, it could mistakenly assume a pending node that was
still a cluster member already fenced, for example SBD was not ready to
function on the peer node yet.

This commit resolves it by getting any existing node cache entry by node
uuid.