While I'd like the tested interface more, it's not implemented
yet leading to a failing test

I mixed up the names when writing the code.  It was always meant to be 'exists'.
No idea how that could ever work.

Note: currently disregards the 'regularity' method.  Will report an error
for test points where the functional is not differentiable.

Functionals do not have to be positive 1-homogeneous, but some important ones are.
For those it is convenient to have a test.

The first non-trivial functional that I tested outside of dune-tnnmg needed
such an epsilon.  The value I used (1e-10) has no particular meaning.
I have no idea what an ideal value is.

Viz, finite difference tests for addGradient, addHessian, directionalSubDiff.
I will delegate writing the actual test to somebody else.

These things can get really complicated, and small errors can go unnoticed
for a long time.

This first patch adds a first test for convexity, and applies it to the
NormFunctional class.