Regret Analysis of Online Gradient Descent-based Iterative Learning Control with Model Mismatch