Learning Interpretable Differentiable Logic Networks for Time-Series Classification