fix(lib): support JSON Schema type arrays in transform_schema (Closes #1876)

[Synxneuos]

Summary

Closes #1876

When a JSON Schema specifies type as an array (e.g., {"type": ["string", "null"]} as generated by Pydantic Optional[T] or Zod .nullable()), transform_schema() previously raised an AssertionError due to unmatched type branches in _transform.py.

Changes

• In src/anthropic/lib/_parse/_transform.py:
  • Added support for list types in transform_schema(), mapping type: [T1, T2] to anyOf: [{"type": T1}, {"type": T2}].
  • Updated type annotations for type_ to Optional[SupportedTypes | list[SupportedTypes]].
  • Allowed is_list(type_) in the terminal validation pass to prevent false assert_never triggers.
• In tests/lib/_parse/test_transform.py:
  • Added unit tests test_type_array() and test_type_array_in_object().

Verification

• All 16 unit tests in tests/lib/_parse/test_transform.py passed (16 passed in 6.03s).

[sylvesterkaczmarek]

Converting a type array from only the type names drops type-specific sibling keywords. For example {'type':['array','null'],'items':{'type':'string'}} becomes an anyOf whose array branch has no items; the original items is only appended to the description, so item type is no longer enforced. The same applies to other type-specific keywords. Could each generated branch be transformed with the applicable sibling schema keywords, or otherwise preserve those constraints?