Returning sparse Jacobians and Hessians

[lambe]

This comes out of my CQP updates in NLP.py. We need to complete the interface returning sparse Hessians and Jacobians from a CUTEst problem and create a version of the CUTEstModel that transforms the return values into Pysparse matrices. The Pysparse version of the model can then be passed directly to solvers like CQP.

[dpo]

And this raises a few other issues:

• [ ] CUTEstModel should return sparse derivatives in coordinate format (rows, cols, vals)
• [ ] subclasses take care of casting that into sparse matrix formats: SciPyCUTEstModel, PySparseCUTEstModel, etc.
• [ ] another subclass can return dense derivatives, but hess() should return sparse derivatives by default, and shess() should be removed.

[kenjydem]

I've made some modifications, let me know if it works.

[lambe]

See the latest pull request. That gets it working.

[dpo]

@kenjydem Please use pull requests to make such changes, so others can comment before code is merged. Thanks.