Enhancing Fractional Gradient Descent with Learned Optimizers