WIP, experimental: Try merging small cells into neighbouring cells

[bendudson]

Reads a mask from the mesh file. Cells that are marked are merged with their neighbours. Takes into account cell volume in the averaging.

• If option "merge_cells" is true, read a field merge_mask from the mesh file
• Cells with merge_mask > 1e-5 are stored in a region
• The time derivatives of N, P and NV in these marked cells are merged with their neighbours in the Y (poloidal) direction: The cell is split in two, each side is averaged with its neighbour, and then the marked cell is the average.
  This takes account of the volume of the cell, so e.g. particle content should be preserved.

[mikekryjak]

@bendudson thanks for this. There is one snag - I think you wrote this branch off afn-neutral-limiters, i.e. the new redoing of the limiters, but I have parked this because we need a better preconditioner to continue. I have switched back to the old limiters implementation in #6.

Because of this, it would probably be more useful to base this PR on master. I haven't done much rebasing before, so for now I just made a new branch (https://github.com/bendudson/hermes-3/tree/merge-cells-master) and cherry picked this commit.