Fix: CTS: handle longer Corosync token timeouts

Previously, startall() would call cluster_stable() immediately after detecting
the "controller successfully started" message. If the Corosync token timeout is
small enough, this will be fine. However with a token timeout of more than
about 1 second, the controllers will not have formed a membership by this
point, causing cluster_stable() to think there are multiple partitions, and
wait for a DC to be elected in each one, when really they will unite into a
single partition in a short time, and only elect a single DC.

Now, startall() waits until seeing that each node is a cluster member before
calling cluster_stable().