fix(utils): keep Optional fields in transform_schema

[MohammadHijjawi97]

Summary

Pydantic v2 emits Optional[X] fields as anyOf: [X, {"type": "null"}], with no top-level type or $ref. transform_schema only handles those two keys. As a result, every Optional field is silently dropped from the simplified schema that GenerateCodeNode, PromptRefinerNode and ReasoningNode pass to the LLM:

class Product(BaseModel):
    title: str
    price: Optional[float] = None

transform_schema(Product.model_json_schema())
# before: {'title': {'type': 'string', 'description': ''}}   <- price missing
# after:  {'title': {...}, 'price': {'type': 'number', 'description': ''}}

Changes

• When a property has anyOf with exactly one non-null variant, describe it as that variant, keeping the field's own description. It then goes through the existing type / array / $ref handling, so Optional scalars, lists and nested models are all covered. Unions of several non-null types are left as they were.
• Added tests/utils/schema_transform_test.py.

Testing

• pytest tests/utils/schema_transform_test.py: fails on pre/beta, passes with the fix.
• ruff check on the changed files; black --check / isort --check-only on the new test file.

Maintainer validation

• Target changed to main at the maintainer's explicit request; the diff remains limited to the two files above.
• Regression test: 1 passed with Pydantic 2.12.5.
• Additional local checks passed for optional booleans, lists of nested models, nested optional integers, unchanged multi-type union behavior, and input immutability. The original bug was reproduced against main.
• Ruff passes on both changed files; Black and isort pass on the new test.
• make lint fails on six pre-existing unused imports in batch_generate_answer_node.py and batch_api.py, both identical to main.
• The full test suite was not run.

[github-actions]

🎉 This PR is included in version 2.3.1 🎉

The release is available on:

• v2.3.1
• GitHub release

Your semantic-release bot 📦🚀