sm::kinematics::quat2AxisAngle and axisAngle2quat are not mutual inverse as much as numerically possible

[HannesSommer]

For example

axisAngle2quat(quat2AxisAngle([1E-9, 0, 0, 1])) == [0, 0, 0, 1] 

but

quat2AxisAngle([1E-9, 0, 0, 1]) != [0, 0, 0]. 

I'm pretty sure this can be improved...

It would be great, because the unit quaternion splines suffer from this that much that I must use ~ eps >= 1E-6 in the finite differences test for the jacobians to get results that are precise not more than 1E-3...

[furgalep]

Ready to close?