Fix: crmd: don't abort transition for expected remote node shutdown

Crmd's match_down_event() now checks the action XML's new "downed" section.
This fixes the following issues with the previous version:

• Stopping a resource that creates a Pacemaker Remote node is now properly detected as a down event for that node. (Stops that are part of a migration are properly ignored.)

• Fencing a cluster node that was running guest nodes is now properly detected as a down event for the guest nodes.

• Unfencing events are now properly ignored.