Support for multiple global grids using a switch function

[Dvegrod]

Added the possibility of supporting several global grids in the same execution by switching the current "focused" grid that the package functions operate with, by using a switch method.

[omlins]

I think this could be done a bit easier by taking an approach as follows:

• remove the library initialization from init_global_grid (means simply to make cuda_functional etc always do the actual check instead of storing it the first time - currently anyway used only once if removed from select_device)
• decompose init_global_grid into init_mpi, create_global_grid::GlobalGrid and set_global_grid
• then expose the new functions create_global_grid and activate_global_grid to the user
• allow update_halo to take a keyword global_grid and finalize_global_grid a positional argument (finalize_global_grid should also accept multiple grids)

Working with multiple grids work then as follows:

1. create grids using create_global_grid
2. activate the grids as needed using activate_global_grid
3. finalize all the grids

It needs to be verified and that the original usage and the error messages remain fully consistent.

Best start a new PR in order to include also the latest code changes from #87