These can be exceptionally challenging while not contributing to the
overall performance. So we might want to skip.

Not yet implemented:

- Vectorization of quadrature loop
- Anisotropic amount of quadrature points per direction

Instationary stuff not fixed so far.

Which play well with the latest change to not assemble on the overlap.

The old formula contained some 'dim's instead of '2's,
which obviously works for 2D....

Not all variants appear in YaspGrids. The number of variant
reduces from 4*d*d -> 2*d for skeleton integrals.