Text this: An empirical comparison of deep learning explainability approaches for EEG using simulated ground truth