4633571a does not change loopsolver constructor for gausseidel

multigridStep uses std::shared_ptr
block size in template of Leaf backend for Gridfunctionspace

input file modSneddon.py for problem taken from Wick [2017]

- make eigenvector computation work for 1D and 3D (damagedelasticenergydensity_decomposed.hh)

  use lapack to compute eigenvectors for all dim, see
  https://gitlab.dune-project.org/core/dune-common/merge_requests/278

- make damagedelasticenergydensitytest.cc work for 1D and 3D
- add eigenvaluetest.cc (now its separated from damagedelasticenergydensitytest)

template parameters have to be reordered with commit 'b93c3ada72875f2b22ec75083b662477bfa3afc2' in dune-solvers modul

Both decomposed and nondecomposed energy implementations can be used.
Can be specified in the input file (decomposition) which implementation should be used.

remove brittle-fracture-memcheck.py

Remove call of python function inside timestep loop in brittle-frcture.cc line 694, 695 and 913
and put it before timestep loop to overcome memory leakage.
But we loose the ability to specify the time dependency of non-homogeneous Dirichlet conditions,
now it is assumed to have linear time dependency.

In LOP we calculate 2*psi, the factor 2 was missing for the damage part.
When calculating the reaction forces we only need the mechanical DOFs (damage=0).

Test for hessian seems to be invalid because FD-approximation of hessian
is not defined due to non continuous gradient.

add simpler output to damagedelasticenergydensitytest

After recent work on dune-tnnmg, the TNNMGStep class does now accept
LinearIterationSteps directly.  The need to insert a
FixedStepLinearSolver is not needed anymore.  Removing it does not
really make the code much shorter, but a bit simpler nevertheless.

The construction of the dune-pdelab GridFunctionSpaces was more
complicated than necessary, to work around

https://gitlab.dune-project.org/pdelab/dune-pdelab/issues/62

This bug has been fixed now, so the brittle-fracture code can be
update accordingly.

not sure if the coupling term is necessary

This avoids messing around with the details of the Python->C++
conversion. Notice that this requires you to update dune-fufem
because make_function is new.

We now provide a single vector-valued function in python.
Splitting into the components is done in c++.