Fix real part of quaternion derivatives (#488)

* Fix real part of quaternion derivatives

In the Taylor expansion of a quaternion rotation, the first-order term should be purely imaginary. Seems like something got translated incorrectly in the change from xpbd to the current integrator

* Update snapshot

* Replace the first-order expansion of rotation with true rotation

This should improve the accuracy of rotations, in particular for faster rotation rates (> approx 0.5 radians per substep for a previous error of 5% in rotation rate)

* Add a .renormalize() method on 3D Rotations

Wasn't sure of the best way to handle this:
* `(self) -> Self` vs `(&mut self) -> ()`
* impl on Rotation vs Quaternion (would be harder because it doesn't belong to Avian)

* Simplify quaternion scalar multiply

Replace the weird thing I was doing with something that should also work in f64

* Update cube simulation snapshot