Fourier-transform-based attribution priors improve the interpretability and stability of deep learning models for genomics