Fix problem in performance of PP algorithm: most of the time was taken by a print that I removed. Now the time of PP is between 1 and 2 ms, which is what I expected.

Implement LP solver using qpOases. Implement a more efficient primal formulation of the LP to compute robustness of a given CoM position (LP2 in the code). Even this formulation sometimes gives wrong results with coin-or solver, but not with qpOases. Implement first performance tests: qpOases seems faster than coin-or solver and fastest formulation seems to be the dual LP (about 0.1 ms for 32 generators). PP algorithm is instead very slow, much slower than in Steve's code, but it's not clear why.

Implement LP solver using coin-or LP. Add in test_LP_solver some small LP tests. Implement primal and dual LP algorithm to compute robustness of a given com position. Test algorithm against PP: primal works always, while dual works only sometimes (need to investigate).