Ugrid save

[pp-mo]

🚀 Pull Request

Description

Beginnings of real save code.

Current status :

• fairly complete initial save code
• simple tests using UGRID cdl example cases
• targetted testing almost complete (...tests/unit/fileformats/netcdf/test_Saver__ugrid.py)
  • still TODO:
    • connectivities with missing points (coming next)
    • optional connectivities
    • edge-based meshes (topology-dim = 1)
    • alternate start-index
    • ... and more ?
• spurious test failures = cartopy coastlines fetch timeout
  • TODO: fix by rebasing, after fixes from 'main' are pulled in to 'mesh-data-model'
  • ?? OR rebase to 'main', if we merge the 'mesh-data-model' feature branch back to main

NOTE :
at present, the first commit is that shown in #4301
Until/unless that itself is merged, it may make more sense to view this relative to the #4301 code

---

Consult Iris pull request check list

[pp-mo]

N.B. this use of separate dims for the last dim of different connectivity types is by design,
but it could seem wasteful.
we could change this, even create all "small" dimension values only once + re-use them

[pp-mo]

Suggested change

# Snapshot the result of saving the example cases.

Copied in error

[pp-mo]

At the risk of some confusion, I'm going to update this with subsequent commits.
Mostly because I fixed some nasty bugs.
However, this is still very much preliminary :

1. I have coded some really basic testcases, and it now passes those, but there is much still to consider
2. the current saver code is now a bit hacky + can probably be tidied somewhat
3. the way this has evolved may have introduced extra unnecessary changes from the preceding code, so that might also be improved. (though could also conflict with overall complexity, i.e. against point (2))

[pp-mo]

I've just done this for now, as it was irritating me + making debugging hard (not to easily see when there is a mesh or not).
This probably belongs in it's own PR.

[pp-mo]

Rebased, should now pass tests.

[pp-mo]

This fixes it so that connectivities of different shapes can compare correctly (i.e. returns False).
Was failing with error, due to problems of array comparison.