expose quadratures before they're scaled by time

[ghorn]

Right now the quadrature interpolation scales by the timestep before interpolation, which is proper. But then sometimes the user will divide the final quadrature state by the end time, like for power optimization. Then we have a T*q/T situation.

If casadi isn't catching this simplification, the sparsity pattern will be denser than necessary. I'm not sure if this is a large effect.