Fix visualisation bug when plot files have ghosts

[mirenradia]

There is a problem with ParaView (and possibly VisIt) when plot files have ghosts. Apparently @KAClough knows what to do...

[KAClough]

Note to self having looked at this a bit more:

• The issue is really one in visit/paraview where it considers the surface through the point(0,0,0) as lying in the boundary cells which are classed as ghost cells. Really this is exactly on the boundary between the domain and the ghosts. When one doesn't write ghosts Visit seems to figure out that this point should be included in the domain.
• It happens both checkpoints (which always output ghosts) and plot files with ghosts.
• Although this is really a visualisation bug it could be fixed by us:

1. It might be possible to just fix it by redefining the valid domain when the data is output.
2. We could only output the boundary cells in sommerfeld directions, and not all ghosts. The ones on symmetric and extrapolating boundaries (which are usually where you try to plot) can be recovered from the symmetry conditions, and the internal ghost cells are clearly not needed at all. This would then reduce enormously the memory footprint of the checkpoint files, so would be worth doing just for that.

[KAClough]

I also note that no one should need to output plot ghosts any more since the boundary updates - which mean that one can set extrapolating bcs to fill sommerfeld cells in any restart from plot files in post processing. So we should consider removing this feature. This doesn't solve the checkpoint issue though.

[mirenradia]

I'm wondering if we can get away with using the GRAMRLevel::m_grown_grids ProblemDomain for writing checkpoints (in the case of Sommerfeld BCs) and not including any other ghosts.

[KAClough]

Yes, that would be great if it works. We would just need to check the chombo copes with reading it back into m_state_new for the checkpoints.