[ci skip]

[ci skip]

Documentation: reorganize README.md and the documentation front page

See merge request !77

It is superseded by the usage hints documented on the front page and by
the linked example project.

- README is shorter, includes only an introduction and links to the full
  documentation for details
- New documented topics: optional dependencies, compiler flags,
  environment variables

- The concept of tnl-bindir is completely useless - it was intended for
  checking if the user configured their $PATH, but the user had to
  already configure  $PATH  in order to use the script.
- The tnl-compile script was used in the quickstart-generated Makefile,
  but gave obsolete output. Moreover, the script approach is inflexible
  and does not provide any benefits compared to hard-coding flags
  directly in the Makefile.

LBM hacks: add optimizations of the ndarray synchronizer as presented at WSC2019

See merge request !76

Advent cleanup

See merge request !75

Moved StaticMatrix from the Containers namespace into Matrices, added matrix-vector multiplication operator

See https://editorconfig.org/ for details.

- pkg-config files are obsolete and not useful anymore, since TNL is now
  a header-only library
- the FindTNL.cmake module was never actually working and it is also
  useless since TNL is a header-only library

The separation is necessary if the external code has a different license
and it is nice even if they have the same license as TNL.

DummySolver was not used at all.

DummySolver was used only in the SolverConfig when calling the
configSetup method - but using an instance of PDEProblem is much
cleaner.

Distributed mesh fix

See merge request !74

DistributedMeshTest: add test with entity centers to ensure that the transferred data really match the physical entities

It depends on input data files, so it is not a unit test.

- iterative way of communicating the global indices for ghost entities
- only one reordering at the end
- a lot of code cleanup

Previously we matched the local subvertex ID with the neighbor's local
superentity ID (entities sharing the same vertex), which does not make
sense. The result cannot use the synchronizeSparse function, so it looks
more complicated.

Note that it still does not work in all corner cases - see the FIXME
notes.

distributeSubentities: fix for decompositions where multiple ranks own all subvertices of their local cells