Fix: scheduler: wait for probe actions to complete to prevent unnecessary restart/re-promote of dependent resources

This addresses the issue brought up from:
https://github.com/ClusterLabs/pacemaker/commit/faf44d811e4f5598dae085c61fdef410c8d18882#commitcomment-22262090

Given an ordering chain in a transition graph like:

A.probe -> A.start -> [...] -> B.start

, if B was already started, it would be scheduled to restart.

Previously, B would be directly stopped, which could turn out to be
unnecessary if A was probed being already started as well. Such
unnecessary restart could be very expensive for heavy workload.

With this commit, a new order will be created:

A.probe -> B.stop

So that any potential restart of B will wait for A.probe to complete. In
case that A is already started, transition will abort and restart of B
won't need to be performed any more.

Similarly for an ordering chain like:

A.probe -> A.start -> [...] -> B.promote

A new order will be created to prevent unnecessary re-promote:
A.probe -> B.demote