Add sparsity functionality to the Jacobian and Hessian functions?

This would be the arguments:

1. colorvec for the color vector of the independent directions
2. sparsity for the sparsity pattern used in the decompression
3. output, jac_prototype, hes_prototype, etc.: the matrix type to be used for the outputted matrix. If mutation then it's just the user given one.

The reason why output can be different from sparsity is because there are many cases where a matrix may be not dense enough for sparse LU-factorization to be efficient, but sparse differentiation may still be a substantial drop in the compute time.

Start by reviewing the Jacobian and Hessian function entry points and the proposed colorvec, sparsity, output, jac_prototype, and hes_prototype arguments. Clarify the API and matrix-type behavior with maintainers, then add tests showing sparse differentiation and the intended output and decompression behavior.