Resolve "multirhstest fails to compile when arpack++ is installed"

Closes #40

See merge request core/dune-istl!161

[cmake] Add description and purpose to feature summary

See merge request core/dune-istl!162

Use structs instead of class for friend declaration

See merge request core/dune-istl!163

Fix segmentation fault in AMG if matrix of one rank is empty.

See merge request core/dune-istl!159

For a matrix with no rows the matrix (built in row mode)
would otherwise never marked as built as this happened only
in the operator++ applied to the last row.

For a matrix with any rows in a parallel run, The subgraph
containing only vertices associated with matrix rows owned by
the process would have 1 vertex although the actual matrix graph
would have none. This is fixed by checking the number of vertices
of the outer graph and act correctly for zero vertices.
For this we need to mark a matrix without rows built in row_wise mode
as built once the CreateIterator is constructed and correct the number
of vertices for a Subgraph of a MatrixGraph with 0 vertices.

Resolve "InverseOperatorResult::clear() does not reset condition_estimate"

Closes #39

See merge request core/dune-istl!158

[multirhstest] Do not use AlignedAllocator unless needed.

See merge request core/dune-istl!157

The `AlignedAllocator` was used for the matrix, however the matrix does not
acually contain SIMD types, so alignment is unessary.

[bugfix] make sure rows are also constructed

Closes #34

See merge request core/dune-istl!155

the row vector is obtained unitialized from an allocator. We now make
sure, that the constructor for all entries is called immediatelly after
allocation.

fixes #34

[Warnings] Fix set-but-unused variables in anisotropic.hh's setBoundary().

See merge request core/dune-istl!124

Deprecate SeqPardiso and remove test_pardiso.

Closes #17

See merge request core/dune-istl!152

The preconditioner `SeqPardiso` using the sparse direct solver
library Pardiso has not been actively maintained for several
years. It is still based on a very old Pardiso version; finding
the library never went into the CMake build system. Moreover,
Pardiso is not open source. With SuperLU and UMFPack, two working
alternative direct solvers are available.

Drop the support for Pardiso by deprecating `SeqPardiso` and
removing the corresponding test.

call `abort()` instead of `assert(false)`

See merge request core/dune-istl!151

remove private `to_real` member function

See merge request core/dune-istl!150

`abort()` will always abort the program instead of silently ignoring
the error when built with `-DNDEBUG`

It reimplements `std::real` and is not used anywhere.

typo: alignement → alignment [ci skip]

See merge request core/dune-istl!149

add changelogs from release to master

See merge request core/dune-istl!148

use `std::make_unique`

See merge request core/dune-istl!147

[test] Add check for BCRSMatrix::nonzeroes()

See merge request core/dune-istl!93

added empty changelog

See merge request core/dune-istl!145

[solvers.hh] Always convert the reduction to double explicitly.

See merge request core/dune-istl!138

This already happened for most solvers when setting the reduction in the
result struct.  It is may be needed when the field type is some extended
number type that prohibits implicit conversion to `double`.  This commit
brings the remaining two solvers in line that did not do this yet.

Fix the #include mess in solvers.hh

See merge request core/dune-istl!137

[preconditioners] fix type of CGSolver reduction parameter after recent updates

See merge request core/dune-istl!132

[bugfix] ensure nonzeroes always return the number of nonzero entries

Closes #28

See merge request core/dune-istl!127

[Warnings] Fix MPI 1 deprecation warnings (by switching to MPI 2).

See merge request core/dune-istl!125