DuneMPI: add spaces around linker flags

See merge request !503

[Ninja] Fix build system self-destructing on clean

Closes #119

See merge request !501

Same reasoning as 4063acee applies,
however there is no LINK_OPTIONS that could be used instead of
LINK_FLAGS [1].  So just add some spaces to make sure that distinct
options aren't joined by accident.

  [1]: https://gitlab.kitware.com/cmake/cmake/issues/16543

List generated files including `CMAKE_CURRENT_BINARY_DIR` in
`DUNE_INSTANCE_GENERATED`.  This means we do not need to set the source file
property `GENERATED` on the generated files, which is only meant for files
generated during build time (e.g. with `add_custom_command()` or during
generate time (e.g. with `file(GENERATE)`).

Closes: #119

DuneMPI: append flags to `COMPILE_OPTIONS` instead of `COMPILE_FLAGS`

See merge request !499

`COMPILE_FLAGS` is a string, but `set_property(... APPEND ...)`
appends to a list.  If the list of compile flags is not empty this
results in a superfluous semicolon in the compile flags which will
break the compiler invocation.

For example, when `MPI_DUNE_COMPILE_FLAGS=-lpthread`, calling
`add_dune_mpi_flags` twice for the same target would result in the
compiler command containing `-lpthread;-lpthread`.

This patch makes use of the `COMPILE_OPTIONS` property instead which
is a list.

address more gcc 8 warnings

See merge request !497

This addresses the following warning from GCC 8:

    dune/common/test/parameterizedobjectfactorysingleton.cc:21:29:
    warning: unnecessary parentheses in declaration of ‘init’ [-Wparentheses]

The `ArithmeticTestSuite` is also used for `bool`.  This causes
several warnings from GCC 8 about using integer operations on bool
such as

    dune/common/debugalign.hh:175:65: warning:
    ‘~’ on an expression of type bool [-Wbool-operation]

    dune/common/test/arithmetictestsuite.hh:683:22: warning:
    ‘<<’ in boolean context, did you mean ‘<’ ? [-Wint-in-bool-context]

This addresses the following compiler warning from GCC 8:

    warning: catching polymorphic type ‘class Dune::RangeError’
    by value [-Wcatch-value=]

[cmake] Add a DUNE_PYTHON_FORCE_PYTHON_VERSION variable

Closes #118

See merge request !485

Extended SIMD interface

Closes #84

See merge request !193

# Conflicts:
#   dune/common/simd.hh

Implement allocator comparison.

See merge request !494

See https://en.cppreference.com/w/cpp/concept/Allocator.  This is required to
use `AlignedAllocator` in a `std::vector`.

make Power functor constexpr

See merge request !493

Fix some warnings (-Wshadow, -Wpedantic, ...)

See merge request !458

Previously the vector-scalar tests assumed that a vector-vector operation was
available and would use that to check the result.  This fails for certain
vectorclass types.

This is a bigger rewrite because I could not understand my own code from a few
months ago.  Hopefully the documentation is better this time...

This will allow me to suppress checking SV or VS operation against the VV
operation if the VV operation does not exist.

This brings us back below 1G.

Constant rvalue-references don't really make sense.  Though I wonder whether
they may arise accidentially in some circumstances.  Anyway, I need to cut
down compilation times somehow...

It's been supseded by `autoCopy()`.

These operators already had shaky support in Vc.  Vectorclass does not support
them at all.

This is done to accomodate vectorclass.

Specifically, we dont require /, %, or shifts with non-scalar rhs.  This is
done to accomodate vectorclass.

We still only allow this if the content matches what was generated previously.

[tuples.hh] Remove, deprecated in 2.6.

See merge request !484

[shared_ptrtest.cc] Remove fallback for compilers without nullptr

See merge request !483

[std/noexcept.hh] Remove, deprecated in 2.6.

See merge request !482

# Conflicts:
#   dune/common/std/CMakeLists.txt

[dune_list_filter()] Fix copy-n-paste error.

See merge request !491