Modifies pollEvents to return latest_GCI as NDB_FAILURE_GCI (~(Uint64)0)
    in case a cluster failure has been detected. An internal flag is set
    in NdbEventBuffer::report_node_failure_completed and the flag is
    reset when the next SUB_GCP_COMPLETE_REP signal is received.
    Function Ndb::isExpectingHigherQueuedEpochs is added to be used together
    with pollEvents2 that checks if cluster has disconnected due to failure
    causing no more events to be received.
    Test case StallingSubscriber is extended to verify that NDB_FAILURE_GCI
    is returned and that polling of events is resumed after the cluster
    is connected again and new epochs are received.