Consider representing hyperelastic materials by displacement gradient

[Andlon]

In #62 we introduced a stopgap design that enables us to implement hyperelastic materials in terms of $\nabla u$ without forming $\vec F$ first. This allows us to avoid the loss of accuracy inherent in forming $\vec F$ when $\nabla u$ is small. However, this design is quite clunky. Long term we might want to always represent hyperelastic materials by displacement gradient $\nabla u$ instead of the deformation gradient $F$. This requires some effort in porting all the materials, and especially in rewriting them to use more numerically stable implementations.