MAINT: improver save refactor and support for overriding all arguments

[cpelley]

This change retains support for existing use of save while aligning it with iris.fileformats.netcdf.save
Enables projects which utilise this function to override the chunking in particular (motivated by EPP).

• save_netcdf now passes kwargs to the underlying iris.fileformats.netcdf.save function.
• Total refactor of the function as there were significant pre-existing issues:
  • Derivation of chunksizes is now constrained properly without making assumptions of which dimensions the grid are mapped to, or assuming consistent grid shape between cubes within the CubeList.
  • Constrained to dimension grid coordinates.
  • save_netcdf call permutations missing from testing are now tested.
  • compression_level keyword argument from save_netcdf still supported and takes precedence but is deprecated in order to better align with its name from iris.fileformats.netcdf.save (i.e. complevel).
  • with_output from improver/cli/__init__.py has been updated to call save_netcdf with complevel but compression_level keyword argument has been kept the same here so as not to break existing workflows that are reliant on the cli (clize, paraflow).

[Copilot]

Pull request overview

This PR refactors improver.utilities.save.save_netcdf to better align with iris.fileformats.netcdf.save, primarily by forwarding arbitrary keyword arguments and documenting where IMPROVER’s defaults differ (eg compression/chunking defaults).

Changes:

• Reworks save_netcdf to accept **kwargs for forwarding to iris.fileformats.netcdf.save, including chunking overrides.
• Introduces argument-normalisation for compression (compression_level deprecation in favor of complevel) and for zlib / shuffle defaults.
• Uses as_cubelist to standardise cube/cubelist input handling.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[cpelley]

Putting this back into draft for the mo. @JoshuaWiggs
The existing chunksize calculation is making wide assumptions.
Making more changes and I don't feel comfortable to leave at it was.

[cpelley]

Nearly there. Interestingly just dealing with an erroneous segmentation fault in the unittests...

=== 7829 passed, 793 skipped, 1 xpassed, 3481 warnings in 208.39s (0:03:28) ====
Fatal Python error: Segmentation fault
Current thread 0x00007f7fdf075440 (most recent call first):
  Garbage-collecting
  <no Python frame>
/home/runner/work/_temp/69157979-fe09-48e0-af0b-b6ae4177db62: line 2:  2575 Segmentation fault      (core dumped) python -u -X faulthandler -m pytest -p no:faulthandler

[Copilot]

Pull request overview

Copilot reviewed 6 out of 6 changed files in this pull request and generated 9 comments.

Comments suppressed due to low confidence (1)

improver/utilities/save.py:109

• save_netcdf's new parameters (zlib, shuffle, chunksizes) are positional by default, which can silently mis-route existing positional calls from the old signature (eg a previous least_significant_digit positional arg would now be interpreted as zlib). Consider making all parameters after complevel keyword-only to avoid accidental positional misuse.

    complevel: int = 1,
    zlib: bool | None = None,
    shuffle: bool = True,
    chunksizes: tuple | None = None,
    **kwargs,