Solver satisfying Dale's principle.

[Seanny123]

Based off the code described in this Nengo issue. Equivalent in effect to the Parisien transform, but has better performance.

Still to do:

• [ ] The L2 regularisation is definitely being done wrong and should be fixed
• [ ] According to @astoeckel, the nnls solver in SciPy has problems and cvxopt should be used instead, as shown in this forum post.
• [ ] The Parisien transform should be included as well as a point of comparison.