Factorize duplication between trltog and trgtol (cpu only)

[dhaumont]

This PR is a refactoring to eliminate duplicated code between trltog and trgtol in the cpu branch.

The duplicated codehas been extracted into new routines in a new module trgl_mod, and they are called from trltog and trgtol.

[dhaumont]

@samhatfield All your suggestions have been pushed. Looking forward to your PR into my branch

[samhatfield]

Unfortunately it wasn't possible to collapse TRGTOL and TRGTOL_COMM into one subroutine, or at least I couldn't find a way (same issue with TRLTOG). The issue is that the stack versions of ZCOMBUFS/R have a shape which depends on YDBUFS%INSEND, and this must first be defined in TRGTOL. So, you have an array whose dimensions must be known at compile time, but the dimensions are only defined at run time...

[samhatfield]

Overall I think this is a nice change and I like how the communication routines are much shorter and the complicated index calculations are done elsewhere. Now TRGTOL/TRLTOG are mostly about communication which makes them easier to read IMO. I'd be happy to merge this, but I'd like to hear an opinion of @RyadElKhatibMF first. I also think we should take this chance to comment the code a bit more.

[RyadElKhatibMF]

yes, it looks fine.

[samhatfield]

I'll make one last check and then I think this is ready to merge.

[samhatfield]

I realised after I started testing that there is still a bug in the checksum dumps for spectral fields. I'd like to test trltog_clean properly looking at checksums etc., but this isn't possible until the bug is fixed, so I opened a new PR here. Let's review and merge that first, then we can merge develop into trltog_clean so we can properly test it?