Fix initialization of MatrixFreeFunctions::ShapeInfo for many quadrature points

[kronbichler]

As pointed out in #7687, our Lagrange polynomials underflow in case we have more than a few hundreds of points. We should take one of the following three steps:

• Equip polynomials with variable-length evaluation, i.e., break the Lagrange polynomial sequence (x_i-x_0)(x_i-x_1)...(x_i-x_n) into several parts if we use more than 100 points or so. It would add quite some extensive logic to what is supposed to be a simple problem. If nobody sees a need for polynomials of such a high degree, I would not like to do this.
• Disable constructing the collocation polynomial if the number of quadrature points is very high, and make sure that we do not use this path anywhere in FEEvaluation.

[masterleinad]

I was discussing this problem with @stefanmeggendorfer and would also prefer the second solution.