fail early on wait_for_master_to_be_ready

[jafreck]

In cases where the cluster will not recover from failure (master node not selected and all nodes in unusable or start_task_failed) the method should fail early. Currently, it will just block until timeout since it assumes the master node selection will be functional.