Text this: Complementary Error Patterns Between Human Evaluators and GPT-4o in Video-Based Cardiopulmonary Resuscitation Skills Assessment