Towards Adversarially Robust Continual Learning