pre-dpnn

This tag will preserve the current functionality for existing users, which I will do my best to continue to support.

The next release will make use of dpnn, which will deprecate all dp.Models and dp.Losses such that nn.Modules and nn.Criterions can be used directly in dp.Experiments and dp.Propagators.

dpnn merges most of the dp.Models and dp.View functionality directly into nn.Modules. I am doing this because I think that dp.Models adds an unnecessary layer of complexity to nn.