Rethink nullspace tutorials

[jorgensd]

Direct solver
PETSc does magic after the fact with nullspace information, which is "correct", but mumps might fail if
icntl_24:1 and inctl_25: 0 is not set for more complex spaces.

Iterative solver
Hypre/Boomeramg ignores the near nullspace if
"pc_hypre_boomeramg_nodal_coarsen": (1-6)
"pc_hypre_boomeramg_vec_interp_variant": (1-3)
is not set.

Thus currently has no effect.

Should consider a two stage approach, stating that constant nullspace is "trivial", and make another one for elasticity,
following guidelines form either:
https://colab.research.google.com/github/firedrakeproject/notebooks/blob/main/03-elasticity.ipynb
or
https://docs.mfem.org/4.8/hypre_8cpp_source.html#l05296
for option tests.

Since the problems are symmetric, Im not sure why I chose gmres and not cg ...

Discussed with @MariusCausemann

[jorgensd]

If you provide a near null space to your matrix with MatSetNearNullSpace() it is ignored by hypre’s BoomerAMG UNLESS you also use the following two options: -pc_hypre_boomeramg_nodal_coarsen -pc_hypre_boomeramg_vec_interp_variant

https://petsc.org/release/manualpages/PC/PCHYPRE/#pchypre