Support functions with a sparse Jacobian

[mohamed82008]

We can support functions with a sparse Jacobian using a struct SparseJacFunc which wraps a vector-valued function. When constructing this function, the sparsity pattern can be input by the user or inferred using ModelingToolkit then an adjoint rule can be defined using this sparse Jacobian and SparseDiffTools. This can be its own package or can live in SparseDiffTools.

[mohamed82008]

This is already implemented in NonconvexUtils and documented in the Nonconvex docs https://julianonconvex.github.io/Nonconvex.jl/stable/gradients/sparse/.