Why is EquelleRuntimeCPU::boundaryFaces() so complex?

[johanseland]

The implemention of EquelleRuntimeCPU::boundaryFaces() is dependent of the internals of Opm::HelperOps and is impossible to adapt to the separataion of inner and outer boundaries. It's sole benefit seems to be that it knows in advance the number of boundary-faces at the cost of a while-loop nested inside a for-loop traversing over all the faces.

I have made a new implementation in RuntimeMPI that simply checks the face_cells flag for boundary::outer, much like EquelleRuntimeCPU::boundaryCells) which is work with inner and outer boundaries.

If this implementation is moved into EquelleRuntimeCPU instead, then RuntimeMPI can simply forward it call an we avoid having two implementations of this.