cmip7: restore flag-variable handling dropped by d82d1cf

[JanStreffing]

Fixes #66. This is the single restoring PR I offered there; going ahead with it since it is mechanical and only re-applies previously reviewed code. Happy to convert it to a revert of the relevant hunks instead if you prefer that shape.

d82d1cf ("Improve and Fix Checks", 19 Jul) rewrote plugins/cmip7/cmip7.py from a base predating #49 and #60, so both merged fixes were removed rather than rebased. v2.3.3 was cut two days later and its release notes credit both by name, but neither is actually in the tag:

ref	_is_flag_variable
v2.3.2	absent
v2.3.3	absent (release notes claim it)
v2.3.4	absent
develop	absent
master	absent

What this restores

Both hunks are byte-identical to what was reviewed and merged.

#49 — variable_id fallback in the len(geo_vars) == 0 branch of _get_geo_var. compliance-checker's is_geophysical excludes anything carrying flag_meanings, which is right for QC flags but wrong for CMIP7 region selectors (basin, siline), whose standard_name is region and which are the file's data variable by design. Without the fallback those files report No geophysical variable detected in the file. at HIGH, which is #48 exactly as originally filed.

#60 — the _is_flag_variable helper and the is_flag gate on the [variable.type] check and the _FillValue / missing_value attribute rules. Without it flag-valued variables are checked against data_type = "float" and constant = 1.0e20, which no integer variable can satisfy.

Existence, dimensions, shape and every other attribute rule still apply. Non-flag variables are unchanged.

Verification

Against a real CMIP7 basin fx file (AWI-ESM3-4-2-veg-HR piControl, int32 with flag_values / flag_meanings, variable_id = "basin"):

develop            geo var = None      1 HIGH: "No geophysical variable detected in the file."
this branch        geo var = 'basin'   0 findings

The synthetic reproducer in #59 still covers both paths.

Note

It is worth checking whether anything else merged between 5a327f0 and d82d1cf went the same way. I only looked at the two changes that affect our files.

[JanStreffing]

Closing as a duplicate of #67, which I missed before opening this. #67 does everything here and more: it also restores 8bd1699 (TIME001 subdaily) and adds dec. Sorry for the noise.

Moving my verification over to #67 so it is not lost.