All kernels using this will have to be refucktored, because
__launch_bounds__ has different meaning in HIP than in CUDA. See [1] for
details.

[1]: https://rocmdocs.amd.com/en/latest/Programming_Guides/HIP-GUIDE.html#device-side-dynamic-global-memory-allocation

Mesh tutorial and documentation changes and some refactoring

See merge request !92

The new value is not correct, it breaks e.g. ParallelFor2D and
ParallelFor3D (for which we had disabled tests by mistake). We need to
separate the max values for x, y and z axes, see the linked Gitlab
issue.

This reverts the commit 37632fd9

The Experimental directory does not make much sense - there is a lot of
experimental or just unused code in TNL, e.g. Images or Operators.

- tests moved to the main UnitTests directory
- added missing inclusion of the Algorithms tests directory to
  CMakeLists.txt

It is an implementation of a specific solver, not part of the general
(header-only) library. If there are some useful generic templates (like
operators etc.), they need to be refactored into the generic interface.

This is more suitable location because the build config tags are used
only by the GridTypeResolver and MeshTypeResolver classes.