Fix: scheduler: check bundle remote connection actions properly

The "REMOTE_CONTAINER_HACK" allows bundles to run on Pacemaker Remote nodes
by setting the bundle remote connection's addr parameter to "#uname" and then
substituting it appropriately when needed.

However, the recorded digest is checked against the calculated digest when
checking actions for changed definitions and expired failures. This occurs
before resources have been allocated to nodes, and thus the substitution may
not be possible. This makes the definition appear to be changed, causing
unnecessary monitor rescheduling, and wrongly clearing any fail count present.

Now, we remember the actions that need the hack, and check them after resources
have been allocated.