Make roundtrip checking more precise + improve some tests accordingly.

[pp-mo]

This provides an important tweak to the split-attrs feature branch, which really needs to be shown working with the main-branch code prior to the adoption of split attributes in cube content and loading (as already merged)

[codecov]

Codecov Report

Patch coverage: 94.63% and project coverage change: +0.16% 🎉

Comparison is base (0b54d58) 89.25% compared to head (f859754) 89.41%.
Report is 139 commits behind head on FEATURE_split_attrs.

Additional details and impacted files

@@                   Coverage Diff                   @@
##           FEATURE_split_attrs    #5402      +/-   ##
=======================================================
+ Coverage                89.25%   89.41%   +0.16%     
=======================================================
  Files                       88       89       +1     
  Lines                    22197    22500     +303     
  Branches                  4858     5396     +538     
=======================================================
+ Hits                     19811    20119     +308     
+ Misses                    1641     1636       -5     
  Partials                   745      745              

Files Changed	Coverage Δ
lib/iris/_merge.py	97.88% <ø> (ø)
lib/iris/analysis/_area_weighted.py	86.89% <ø> (ø)
lib/iris/analysis/_interpolation.py	95.80% <ø> (ø)
lib/iris/analysis/_regrid.py	83.33% <ø> (ø)
lib/iris/analysis/trajectory.py	94.44% <ø> (ø)
lib/iris/common/metadata.py	93.23% <ø> (ø)
lib/iris/common/resolve.py	92.88% <ø> (ø)
lib/iris/config.py	81.17% <ø> (ø)
lib/iris/experimental/regrid.py	77.38% <ø> (ø)
lib/iris/experimental/ugrid/mesh.py	92.18% <ø> (ø)
... and 30 more

... and 3 files with indirect coverage changes

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[pp-mo]

Closing pending a re-submission.

I'm not happy with the apparently spurious commit from pre-commit-ci aiming to fix what is already fixed elsewhere (i.e. "later") on the feature-branch.
And similarly the test failures from lib/iris/tests/test_coding_standards.py::test_netcdf4_import due to direct use of the netCDF4 module -- also a problem fixed "later" in the feature-branch.
So we really don't want to fix those problems here since it will probably create conflicts.

Following discussion with @trexfeathers , I think we can simply logically accept that, from results in #5403, these proposed changes to the "status-quo" tests are valid.
So I will rebase this change onto the current head of the target branch, and include it in a forthcoming "netcdf split attributes saving PR" , as-per #4986