support for forward automatic differentiation

[martinResearch]

Thanks you for sharing this library. In some case using forward automatic differentiation can be useful, for example when solving non-linear least square problems, where having the jacobian matrix of the residuals through auto-diff allows to use the fast levenberg-marquardt method. It would be great to add support for forward automatic differentiation in MyGrad. Is that something that could possibly be added to the roadmap? If useful as a reference, I wrote this Matlab library https://github.com/martinResearch/MatlabAutoDiff that stores the gradient as sparse matrices.

[martinResearch]

I just found this library https://github.com/PTNobel/autodiff that does forward auto diff an support sparse jacobians.