Implementation of meshing for spherical shell

Hi,

when the mesh generation for the spherical shell was ported from HHG to HyTeG, only the on-the-fly variant for discretisation of the unit sphere was selected. This is a little problematic, as this variant does not use midpoint refinement and introduces a certain bias in the mesh.

We should also port the "classical" approach based on midpoint refinement. This also has the advantage that also the coarse meshes generated for different ntan = 2^k+1 values are nested.

Cheers
Marcus