Add sparse routines sobj and sgrad

[jfowkes]

Resolves #73

The official CUTEst MATLAB interface has the sparse routines:

Tool    CUTEst function   Purpose
--------------------------------------------------------------------------
sobj    uofg / cofsg      Evaluate objective function value
                          and its sparse gradient if requested
sgrad   ugr / cisgr       Evaluate objective function or
                          constraint gradient as a sparse vector

These are the sparse equivalents of our existing obj and grad routines.

This is an attempt to implement these in a similar way as done in MATLAB: https://github.com/ralna/CUTEst/blob/master/src/matlab/mcutest.c

Major caveat: CUTEst only supports sparse gradients for constrained problems, so we return dense gradients for unconstrained problems as in the MATLAB interface and convert them to scipy.sparse.coo_matrix.

• [x] Add C interfaces for sobj,sgrad to c_interfaces.py
• [x] Add Python interfaces for sobj,sgrad to problem_class.py
• [x] Write comprehensive unit tests for sobj,sgrad
• [x] Write documentation for sobj,sgrad

[jfowkes]

There is a segfault on lines 458-461 of cisgr.F90, looks like an upstream bug to me.

[jfowkes]

There is a segfault on lines 458-461 of cisgr.F90, looks like an upstream bug to me.

This was indeed an upstream bug, and has now been fixed in CUTEst.

[jfowkes]

@lindonroberts this is now ready for you to review, should be a simple one.

[lindonroberts]

Looks good to me!