DeepAREstimator: Elaborate on the docs for nonnegative_pred_samples and provide a way to force DeepAR predictions to be positive

[Serendipity31]

The addition of the option to set nonnegative_pred_samples = True in the DeepAREstimator, is great! However, it's not clear how this relates to, or interacts with, the choice of output distribution. It would be really useful if the docs could address this point, perhaps by expanding also on the role the distribution output plays in prediction process.

Additionally, it would be really useful to be able have a similar boolean that would allow the user to specify that all predictions must be positive (and not just non-negative).

In my case, the series I'm modelling take strictly positive values. Setting `nonnegative_pred_samples = True' does remove the possibility of negative prediction samples, but does return the value 0.0 sometimes.