High-dimensional Analysis of Knowledge Distillation: Weak-to-Strong Generalization and Scaling Laws