Add closest point projection based on Projected Gradient with a Simplex projection algorithm

[jorgensd]

After discussing with @thomas-surowiec we decided that the fastest approach is probably the projected gradient method.

Using the algorithms from the following papers:
Simplex projection:

• Algorithm 1 in Laurent Condat. Fast Projection onto the Simplex and the l1 Ball. (2016) DOI: 10.1007/s10107-015-0946-6, which is from:
  • Held, M., Wolfe, P., Crowder, H.: Validation of subgradient optimization (1974) DOI: 10.1007/BF01580223

Using the Goldstein-Levitin-Polyak gradient projection method, with linesearch method from:

• Dimitri P. Bertsekas, "On the Goldstein-Levitin-Polyak gradient projection method" (1976) DOI: 10.1109/TAC.1976.1101194

Implements reference solver using scipy, a python implementation for debugging, and a highly optimized C++ version that allows for threading.