When using transaction sticky load balance policies, before making an
invocation, we check whether there's a sticky target in the tx failover
authorisation map and if there is, we add it to the transient payload.

The sticky lbps then check that sticky target and if available in the cluster,
they use it. However, if there's a sticky target was set but it's not available,
we chose a new target when indeed we should fail cos the transaction
already reached the server (if it hadn't reached a server yet, there're
wouldn't be a sticky target at all).

I can see this causing issues when shutting down a node that was being
used in a transaction. The problem is that the node shutdown will be chosen
as sticky target but as it's not available, so it choses a a brand new one. It
should have instead failed.