Entropy-Kantorovich potentials for training CNFs

[ChrisRackauckas]

https://arxiv.org/pdf/2006.06033.pdf from @cfinlay would be interesting to add as a layer in contrast to the FFJORD layer. It could train using the differentiable adaptive Monte Carlo from https://github.com/SciML/Quadrature.jl and then be setup so that the forward pass uses the ODE solver. This would just take an @adjoint definition so that the gradient calculation does something very different.