resolve outstanding test failures in chardata handling code #6898

[pp-mo]

Additional to #6898
As-of 2026-01-29
There are still quite a lot of test failures
~104 : https://github.com/SciTools/iris/actions/runs/21443853740/job/61753974741?pr=6898

But from a quick scan, I suspect a lot of these (if not most) may be unit-test mock type problems.
I E. the type of tests which tends to be "fragile" when refactoring an implementation.

Hopeful or indicative signs:

• Unit-tests for encoded datasets are passing (but may need extending)
• basic netcdf read+write integration tests for string data are passing
• plenty of other existing tests, which rely on netcdf load or save, are not failing

[pp-mo]

Update 2026-03-06

Various test fixes now applied to #6898, at this point.
Largely just updates to mock-ist tests

However there is an outstanding problem, which is that we have a test file with a character (label) variable where the string dimension is not its final dimension :
This is the file used by iris.tests.test_cf.TestLabels, at NetCDF/label_and_climate/A1B-99999a-river-sep-2070-2099.nc,
i.e. here

From this account in the latest CF (v1.13), we have

If the variable has a type of char, the inner dimension (last dimension in CDL terms) is the maximum length of each string, and the other dimensions are axis dimensions.

So, I think this isn't really good CF. Though that probably wasn't clear in early versions of the CF spec.
But this probably needs discussion.

In practice, to retain existing behaviour we would have to follow links between variables, which is inevitably in the CF domain and can't easily --or even shouldn't-- be done in the 'EncodedDatasets' layer