Merge branch 'expression-templates' of mmg-gitlab.fjfi.cvut.cz:tnl/tnl-dev into expression-templates

This reverts commit a911332c.

Running 2 jobs concurrently instead of 8 threads per job is much better.

Calculation of bandwidth broke in commit fb97bcd0

Check of MpiCommunicator::IsInitialized is necessary because
non-distributed benchmarks like tnl-benchmark-blas do not initialize
the communicator.

It does not have any effect on recent builds and we explicitly set
-march=native -mtune=native later in the file.

Distributed linear solvers

See merge request mmg/tnl-dev!8