Jacobian computations are inexact on curved meshes

[jlchan]

Currently, we compute J at interpolation points. This is exact when J is a low enough degree polynomial, but in general it is degree d(N-2) in d dimensions so this approximation may not be exact on curved isoparametric meshes.

This probably isn't a big issue now, but if exactness of the global quadrature wJq is needed, this can be problematic.