Efficient Long-Term Prediction of Chaotic Systems Using Physics-Informed Neural Operators: Overcoming Limitations of Traditional Closure Models