OfflineData: 3D support: normal and normal mass may not be computed correctly across MPI ranks

[tamiko]

The logic in OfflineData<dim, Number>::construct_boundary_map currently reads as follows:

for (auto cell = begin; cell != end; ++cell) {

  if (!cell->is_artificial() /* !cell->is_artificial_on_level() /*)
    continue;

  for (auto f : GeometryInfo<dim>::face_indices()) {
    const auto face = cell->face(f);
    const auto id = face->boundary_id();

    if (!face->at_boundary())
      continue;

    // ...

    /* Skip nonlocal degrees of freedom: */                                
    if (index >= n_locally_owned_)                                         
      continue; 
  }   /* f */
}     /* cell */

This logic only works in 2D and will fail in 3D when we encounter hanging nodes on boundaries. We might also miss to set up boundary degrees of freedom.

[tamiko]

Fix should be straightforward: We need to assemble the normal (and boundary mass) into a temporary vector with AffineConstraints::add_local_to_global() and compress(VectorOperation::add) after assembly.
Then, we can collect the normal and put it into our BoundaryMap.