When using the tpfa laws in an mpfa context, scvf.center() gives
the centers of the half faces, thus, a worse tpfa approximation.
We use ipGlobal() here now, so in an mpfa context one would have to
use the standard ipGlobal() which is the facet center. Then, flux
comutation is guaranteed to be of standard tpfa-type. An overhead
remains as it is calculated for each half face.