F-Fidelity: A Robust Framework for Faithfulness Evaluation of Explainable AI