Refactoring the Lax-Fridrichs scheme for the transport equation.

 - sign
 - Heaviside
 - smooth Heaviside
 - shift
 - rotation

Curiously clang complained about it, but nvcc didn't.

The original algorithm used space as one of the separators between types
in template parameter lists, so as a result "A< short int >" was parsed
into a list of ("A", "short", "int").

This prevented scaling of GMRES on small grids (up to 240x240)!

Merge branch 'hamilton-jacobi' of geraldine.fjfi.cvut.cz:/local/projects/tnl/tnl into hamilton-jacobi

Conflicts:
	src/TNL/Containers/Array_impl.h
	src/TNL/Meshes/GridDetails/Grid1D.h
	src/TNL/Meshes/GridDetails/Grid1D_impl.h
	src/TNL/Meshes/GridDetails/Grid2D.h
	src/TNL/Meshes/GridDetails/Grid2D_impl.h
	src/TNL/Meshes/GridDetails/Grid3D.h
	src/TNL/Meshes/GridDetails/Grid3D_impl.h
	src/TNL/Solvers/PDE/TimeDependentPDESolver_impl.h

We must pass the CUDA_HOST_COMPILER variable manually to use the
compiler configured by the user.