Regularization-Based Efficient Continual Learning in Deep State-Space Models