Due to a bug in dune-geomtry the volume calculation is not accurate enough. For
now we just remove all volume calls from the weak formulation.

Can me generated by calling generate_meshfiles.py

Have to investigate further. The change in the residual seems way too
large. The curious thing is that the input to the sum factorization kernels
looks good and the poisson without sum factorization gives the same
solution. Very unexpected.

There was a minus sign error...

Instead of generating gmsh files for a structured grid we use a real
unstructured grid.

The residuals of these tests should be the same modulo permutation and fuzzy
float compare.

Note:

- direct_is_possible true/false could probably be handled in an upper/lower
  vectorization way.

- Vectorization of SF kernels should be based on cost permuted matrix sequence.

There seems to still be some problem.

Note: Not yet working for gmsh files