Segregate XTCE 1.2 and 1.3 test expectations with a version-aware suite scheme

[medley56]

Follow-up from review of #291.

#291 added XTCE 1.3 support, which means the test suite now carries expectations for two versions of the standard. A partial split landed in that PR — version-specific tests were moved out of test_definitions.py and test_validation.py into modules with a stated remit:

module	remit
tests/unit/test_xtce/test_versions.py	the version registry itself
tests/unit/test_xtce/test_xtce_1_3_features.py	constructs that exist only in 1.3
tests/unit/test_xtce/test_version_conversion.py	detection, retargeting, version-dependent serialization
tests/unit/test_xtce/test_version_validation.py	version-dependent validation

That stops the interleaving, but leaves the mechanical half of the scheme undone.

Remaining work

1. A version-parametrized fixture in tests/conftest.py with params=SUPPORTED_XTCE_VERSIONS and ids xtce12 / xtce13, so behavior that should be identical across versions is asserted once and runs per version, rather than being written out twice or tested against only one.
2. xtce12 / xtce13 markers registered in pyproject.toml, applied via module-level pytestmark, so -m xtce13 selects the 1.3 surface.
3. Version-specific test data under tests/test_data/xtce_1_3/, moving test_xtce_1_3.xml and test_xtce_1_3_only_features.xml there. Note this interacts with the drift-guard test added in Add support for XTCE 1.3 alongside XTCE 1.2 #291 (test_xtce_1_3_fixture_tracks_the_1_2_fixture), which pins test_xtce_1_3.xml to being a pure version-swap of test_xtce.xml — the guard should move with the fixture.
4. Keep test_encodings.py version-neutral. It is currently close to that, but the encoding-level tests for the 1.3 delimiter-only string form should be checked against the boundary.
5. Consider whether test_version_conversion.py and test_version_validation.py should merge once the fixture exists, since some of what separates them today is which module the moved tests came from.

Why not in #291

The PR was already ~8k lines across 23 files, and none of the other review findings depended on this. Doing the suite-wide mechanical change separately keeps both diffs reviewable.

🤖 Generated with Claude Code