Reformulation of the parallel Riccati kernel

[edantec]

The current formulation of the parallel solver does not allow to cycle the solver data easily. As a consequence, we must reallocate the LQR subproblem solver data at each control cycle in MPC context.

This issue is related to the solver cycling PR in which a first implementation of cycling for parallel solver is proposed (but does not work yet).