Facet assembly and MappingIsoparametric

[kinnala]

Because I never was a great user of quad/hex elements, the boundary mapping -related parts of MappingIsoparametric do not exist. This means that facet assembly does not currently work at all with quads and hexes.

Fixing this issue requires writing tests and implementing the remaining methods in MappingIsoparametric.

Note that this was already implemented in the Python 2.7 version of the code: https://github.com/kinnala/sp.fem/blob/master/spfem/mapping.py#L173. This gives lots of hints but I hope the new version would use np.einsum like the other mapping methods.

Update 28.11.2018: Write more tests and this can be closed.

[kinnala]

Quadrilateral boundary assembly is now at least partially working in 77bd5fc. Getting the indexing correctly for hexahedral elements is harder than I anticipated and requires more work. MappingIsoparametric.normals is just a stub implementation so that the current version can be tested.

[kinnala]

It seems that the indexing was correct. In the inverse isoparametric mapping you need a numerical method at each quadrature point since the mapping cannot be inverted analytically. It turned out that I had a stupid initial guess, origin of the reference system. In 2D this worked fine (Quads) but in 3D the Jacobian determinant at origin seems to be zero. More tests are required before this issue is closed though.

[kinnala]

Also, normal vector implementation is missing.