Subdomains & boundaries in refinement

[kinnala]

The subdomains and boundaries are currently invalid if one refines the mesh. Create a method to Mesh which removes them on refine and override in MeshTri etc. to create new indexing

Boundaries:

• MeshTri uniform
• MeshTri adaptive
• MeshQuad
• MeshTet
• MeshHex

[gdmcbain]

This would be perfect for the empirical convergence study of #13; thus far I've been remeshing the Gmsh GEO from scratch.

[gdmcbain]

Yes, and this would be even more comvenient in conjunction with the adaptive use of Mesh.refine demonstrated in #80.

[kinnala]

For now I added a warning that the indexing is invalidated.

Subdomains are simple to maintain in refine, but now that I thought about this more carefully I think that maintaining the named boundaries is very hard using the current data structures. There are many edge cases so you have to be very careful or do something clever.

[kinnala]

Uniform refine in MeshTri now should preserve named boundaries (stuff in Mesh.boundaries) in ec297c1. I was overly pessimistic in the previous comment: it was quite straightforward although it required some meditation.

[kinnala]

I realized that there is still a bug in refine when called with integer larger than one. This is just a reminder for myself.