working on the MHFEM section - reorganized tables

- tables include results only on the finest meshes
- different grouping - OpenMP and MPI in the same table, GPU results in
  a separate table
- recomputed benchmarks on the finest 3D mesh with 1-24 cores