Quadrature degree is underestimated for elements where order != polynomial degree

There is a convention in UFL/FFC/FIAT that the order of the element is such that
the finite element space is included in the space of polynomials of degree 'order'.
(This is why we start the Raviart-Thomas's at 1.)

While FFC and FIAT have a distinction between degree() and embedded_degree(),
UFL don't. And for the auto_estimate_total_quadrature_degree, only
the UFL degree is used, making the distinction between degree() and
embedded_degree() essentially irrelevant.

This is unsafe.