Add finer control field methods to read and write halos from accelerated devices

[sergisiso]

Closes #53 by adding parameters to the read_from/write_to_device methods. The halos in the generic exchange are now passed using this finer-grain data selection.

Also:

• Closes field%set_data() should update accelerator devices #29 : Adds write_to_device methods and fixes set_data().
• Added a new test to check the device acceleration functionality (it is not exhaustive yet but at least it starts providing some examples and guarantees)
• Switched to GitHub Actions
• Switched device pointers to type(c_ptr) instead of integer(c_intptr_t). The reason is that the later is exclusive to how OpenCL works (and causes some compiler issues). The c_ptr is more generic and fits better with SYCL and Kokkos.

[sergisiso]

@arporter This is ready for review. It prepares dl_esm_inf to expose more information about the device reads and writes in order to make the MPI+device computation more efficient (and correct!).

I have already use it in manual PSycloneBench and PSyclone to prove that it provides correct results. I will clean up and submit those PR next.

[arporter]

Nice work Sergi, I particularly like the new test. It works out of the box for me.
Just a few very minor things to tidy up.

[arporter]

Please could you add an ONLY clause to this import to make it easier to follow the code.

[sergisiso]

Removed

[arporter]

Not quite :-) I meant changing this to something like, use iso_c_binding, only: c_loc, etc. etc.

[sergisiso]

Ahh, I think I must have replied thinking it was the previous comment, I did not address this suggestion the first time.

Now its done

[sergisiso]

@arporter I made the suggested changes. This is ready for another look.

[arporter]

Almost! Just the one outstanding comment to deal with.

[arporter]

Not quite :-) I meant changing this to something like, use iso_c_binding, only: c_loc, etc. etc.

[sergisiso]

@arporter Ready again, I missed the comment the last time.

[arporter]

All requested changes have been made. New test still compiles and runs OK.