In Shift and In Variance: Assessing the Robustness of HAR Deep Learning Models against Variability