Implementation of CHOLMOD Solver

See merge request !36

This class wrappes the dune-istl CHOLMOD method for dune-solvers module.

It is basically a copy of UMFPack solver.

There are problems with some METIS implementations (especially LibScotchMETIS)
causing random segfaults. Therefore the usage of METIS is disabled for matrices larger
than 3000 rows for now (whichs seems stable),
see https://gitlab.inria.fr/scotch/scotch/issues/8

CI: Disable C++14 runner

See merge request !41

The image of debian 9 with C++14 compiler is not updated any more
and incompatible with the new allowed usage of C++17 features.

UMFPackSolver: Remove several hard-wired linear algebra types

See merge request !40

Removing these assumptions about the linear algebra types
is one step towards support for unnested linear algebra,
like BCRSMatrix<double> etc.  More work for that is needed,
though.

Give ObstacleRestrictor an empty virtual destructor

See merge request !37

Otherwise, non-default destructors in derived class are not called
properly when handling a derived class through a base pointer.

H1SemiNorm: use smart pointer for matrix

See merge request !35

Provide mutable access to matrix in transfer operator

See merge request !34

This is needed to use the new assembleGlobalBasisTransferMatrix(...)
assembler functions which needs access to the internal raw matrix.

Update CI config to clang-7

See merge request !32

The old image is no longer updatet

Other shall not doubt its right of existence in
times of perfect forwarding and convenient cv-modifiers.

Fix whitespaces in ProjectedBlockGS

See merge request !31

There were still some spurious whitespaces left from d49b1348.

init public members with nullptr for more meaningful fault messages if not set

See merge request !30

This matches how to specify search directories for optional
dependencies in DUNE's core modules.

do not catch polymorphic type by value

See merge request !28

CGStep: hold preconditioner in a `shared_ptr`

See merge request !29

The `#include` directives in cgstep.cc otherwise result in definitions
in the wrong namespace.

MultigridStep: throw exception when no smoother is provided

See merge request !27

Otherwise there will be a less user-friendly null-pointer dereference
in `iterate()`.

The CI segfaults for this combination and at least I was not able to
reproduce/debug this.

Use sparseRangeFor utility instead of raw iterators

See merge request !25

Feature/multigridstep rework set smoother

See merge request !26