Wrong num_cells for DistributedTriangulations with ghost cells

[Jai-Tushar]

If the DistributedTriangulation has overlapping host cells, the following code counts some cells several times:

function Geometry.num_cells(a::DistributedTriangulation)
  sum(map(trian->num_cells(trian),local_views(a)))
end

[JordiManyer]

I think in practice this is only used for display functions, but it is certainly a bug. It should be fixed by #160