Note: Right now this is only done for FastDGGridOperator.

Funnily, we were only able to calculate symmetric problems so far.

It is not (yet!) suitable for our needs.

Note: sumfact_poisson_fastdg_gradvec_nonquadvec does not work!

At the moment loopy tries to acces a vector class variable and throws
a not implemented error. Probably needs some changes in the horizontal
add code.

For some reason, the patch refuses to schedule the kernel for sumfact_poisson_dg_symdiff.