Learning Data-Driven Vector-Quantized Degradation Model for Animation Video Super-Resolution