GPU acceleration

[mesonepigreco]

We can rewrite the basic structure of the code in Julia and exploit the complete GPU-parallelizable environment.
This can be achieved like the C++ version of the code, but entirely in julia, to avoid PyCall problems.
The python code could export all quantities like it is done for the C++ version.

This is particularly relevant for parallelization via GPU and MPI hybrid, that could be exploited in this way.