Explaining Graph Neural Networks with Large Language Models: A Counterfactual Perspective for Molecular Property Prediction