A Useful Distribution for Fitting Discrete Data: Revival of the COM-Poisson