Text this: A comprehensive analysis of perturbation methods in explainable AI feature attribution validation for neural time series classifiers