Tiny is not small enough: High-quality, low-resource facial animation models through hybrid knowledge distillation