Reduction of Q for mimic joints

[Stéphane Caron]

We found with @gergondet yesterday an issue in the reduceQC function: cross-terms in alpha_i * alpha_j are missing in the computation of Q.

Mimic joints can be written m_i = alpha_i * q_i + offset_i, so that mdd_i = alpha_i * qdd_i, or in matrix form mdd = A * qdd. Then, the calculation should look like this (z=mdd and x=qdd):

This is a substitution. We should update the function, or convince @aescande to demonstrate the power of tvm here 😉