Refactoring functionals

[rckirby]

It seems like our code would be a lot cleaner if we could have sets of functionals (the gradient at a point, in Cartesian or tangent/normal coordinates, moments against all polynomials of some degree on a facet, etc).

Advantages:

• Easier implementation of elements
• Easier to batch together the evaluations in to_riesz
• Possibly easier inference of zany mappings in FInAT (hypothesized).

Challenges:

• This is a pervasive change in functional.py and may require reimplementing all our Ciarlet finite elements
• We need to make sure not to bump anything indirectly (e.g. code generation of general interpolation operators)