Learning-Augmented Algorithms for the Bahncard Problem