Hyperparameters in Continual Learning: a Reality Check