fix: honour exclude_defaults for link fields

[simontaurus]

Follow-up to #166, which surfaced this while testing .json(exclude_none=True, exclude_defaults=True, indent=2).

Declaring a default IRI for a link already works and is already seeded at construction (__link_defaults__, src/oold/model/_descriptor.py:133):

employer: Link[Org | None] = OoldField(default="ex:default-org")      # v2
employer: Optional[Org] = Field("ex:default-org", range="Org")        # v1

But exclude_defaults=True never omitted it. A link value is routed out of the field pydantic validates, so that field always sits at None and pydantic's comparison answers "not a default" for every link, set or not.

Changes:

• _matches_link_default compares the stored IRI(s) against __link_defaults__, reading both sides as lists so a to-one default seeded into a to-many link still matches
• applied in the v2 model serializer, in the v1 dict() override, and in to_json, which re-added link IRIs after the dump had excluded them
• v1 dict() also wrote an unset link back as None past exclude_defaults, exclude_unset and exclude={...}; v2 already filtered those and v1 legacy omits them, so this is a parity fix in both directions
• three tests in tests/test_serialization.py, parametrised over both versions, importing the descriptor bindings directly because the legacy bindings do not record link defaults at all

740 passed under both bindings.

[github-actions]

Release preview

No version bump from the current commits (stays at v1.0.2). Use conventional commit types (feat, fix, ...) to trigger a release.

Changelog preview (truncated)

_{Preview via python-semantic-release and conventional commits.}

[github-actions]

📊 Benchmark Results

Click to see benchmark comparison

📊 Benchmark Comparison (threshold: 1.3x)
============================================================

➖ Unchanged (within threshold):
  ➖ test_simple_dict_document_store: 0.0019s → 0.0019s (-0.1%)
  ➖ test_sqlite_document_store: 0.0021s → 0.0021s (+0.3%)
  ➖ test_local_sparql_store: 0.0381s → 0.0392s (+2.6%)
  ➖ test_oneof_subschema: 0.0597s → 0.0610s (+2.2%)
  ➖ test_enum_docstrings: 0.0501s → 0.0514s (+2.5%)
  ➖ test_subclass_inheritance: 0.0534s → 0.0552s (+3.5%)
  ➖ test_class_hierarchy: 0.0521s → 0.0520s (-0.2%)
  ➖ test_core[v1]: 0.0352s → 0.0374s (+6.2%)
  ➖ test_core[v2]: 0.0435s → 0.0452s (+3.9%)
  ➖ test_schema_generation[v1]: 0.0016s → 0.0016s (+0.5%)
  ➖ test_schema_generation[v2]: 0.0038s → 0.0038s (-0.1%)
  ➖ test_simple_json: 0.0007s → 0.0007s (+2.4%)
  ➖ test_complex_graph: 0.0016s → 0.0017s (+1.7%)

============================================================
Summary: 0 regressions, 0 improvements, 13 unchanged
============================================================

✅ No significant performance regressions

Threshold: 1.3x (30% slower triggers a regression warning)

Note: Benchmarks are informational only and won't fail the build.

_{💡 Tip: Download the benchmark-results artifact for detailed JSON data}

[codecov]

Codecov Report

❌ Patch coverage is 80.95238% with 4 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/oold/model/_compat.py	66.6%	1 Missing and 1 partial ⚠️
src/oold/model/_descriptor.py	75.0%	1 Missing and 1 partial ⚠️

📢 Thoughts on this report? Let us know!