On the Limitation and Experience Replay for GNNs in Continual Learning