Add Phase 2 end-to-end integration test for native SPDX 3 constructs

[Copilot]

Adds a single combined test for the five recently merged native SPDX 3 features (N1 fragment origin, N2 declared/concluded license, N4 external identifiers/refs, N5 base-model lineage, N6 dataset creator/publishedBy) to catch interaction bugs and residual-annotation duplication.

Changes:

• New tests/test_phase2_integration.py exercising all five constructs on one AI-model input.
• Assertions that native fields are present and correct:
  • ExternalMap round-trips through spdx-python-model.
  • hasConcludedLicense relationship and simpleLicensingText for a scanner-detected license.
  • DOI ExternalIdentifier, arXiv and model-page ExternalRef.
  • descendantOf relationship with stub base-model package.
  • Agent creator and publishedBy relationship from dataset.
• Checks that provenance Annotations do not duplicate values now covered natively (DOI, arXiv, URL, creator, license value).
• Verifies byte-identical determinism across two generation runs.
• Validates the combined document via spdx-python-model deserialization.

Example:

model = AiModelMetadata(
    name="phase2-integrated-model",
    license="mit",
    doi="10.1234/example.doi",
    arxiv_ids=["2301.12345"],
    url="https://huggingface.co/example/phase2-integrated-model",
    base_model="openai-community/gpt2",
    base_model_relation="finetune",
    datasets=[DatasetReference(role="trainedOn", metadata=DatasetMetadata(
        name="ai-training-data",
        creator="Example Research Group",
    ))],
)
exporter = build_model(model, CreationMetadata())

[bact]

dup of #112