Boltzmann convolutions and Welford mean-variance layers with an application to time series forecasting and classification