Unfortunately with gcc-13.2.1 with optimizations there is a warning, which I was not able to correct.

    ...
    inlined from ‘std::vector<std::tuple<long int, std::reference_wrapper<ProcessLib::LocalAssemblerInterface> > > {anonymous}::collectActiveLocalAssemblers(const BaseLib::PolymorphicRandomAccessContainerView<ProcessLib::LocalAssemblerInterface>&, const std::vector<long unsigned int>&)’ at /var/lib/gitlab-runner/builds/F1XUyv4cx/1/ogs/ogs/ProcessLib/Assembly/ParallelVectorMatrixAssembler.cpp:132:62:
/usr/include/c++/13.2.1/bits/stl_iterator_base_funcs.h:175:9: warning: iteration 576460752303423487 invokes undefined behavior [-Waggressive-loop-optimizations]
  175 |         while (__n--)
      |         ^~~~~
/usr/include/c++/13.2.1/bits/stl_iterator_base_funcs.h:175:9: note: within this loop

Other compilers gcc-14, clang-17 and 18, msvc had no problems with the
former code.

This is mostly copy from the VectorMatrixAssembler.

A little bit :) 0.5% maybe. Measured on TRM/Mockup ctest.

This costs some memory (now element ids and local assembler references are stored)
but simplifies the code. If local assemblers would return an element id,
the code could be simplified further.

This slightly simplifies code avoiding extra boolean variable.

This is not sufficient for the staggered scheme yet.

Same as 3e9a70cd but for global matrices.

A little code duplication in favor not to deal with pointers.

A little code duplication in favor not to deal with pointers.

The matrices are not in use by any process.