normalize quaternions and DCMs before using them to rotate vectors

[ghorn]

This will add a Floating requirement, but should avoid some potentially serious problems, especially in optimization code