Supersede SQ Cardinal Points with GWS computation via random sampling, simulation and optimization

[pattacini]

It would be very much appealing to exploit the whole surface of SQ and not just its cardinal points. We can employ the same methodology of #11, where the SQ's surface gets meshed using CGAL for fast intersection and normals retrieval.

Given the mesh and the normals, we take out portions based on some geometrical considerations, and then we can apply random sampling (~ brute force):

1. Put the hand in the pre-grasp configuration.
2. Pick up one random point on the mesh with the associated normal.
3. Align the approach vector of the hand to the said normal.
4. Establish a policy to find out the remaining angle around the approach vector: sample and/or reason.
5. Simulate the approach until a collision is detected.
6. Backtrack to reach a location where there are no collisions.
7. Simulate the hand closure (like in Automatically adapt closing fingers posture to the SQ #11) to get the contact points.
8. Use a nonlinear optimizer (e.g. Ipopt) to evaluate the Grasp Wrench Space (GWS), once a friction coefficient is provided.

[pattacini]

Point 8 could be done also by means of https://github.com/humanoid-path-planner/hpp-centroidal-dynamics.