Fix #7203 - Preserve type info for non-object schemas in json_to_pydantic

[Whning0513]

schema_to_pydantic_model loses $defs type information (array -> object)

When converting a JSON schema generated by Pydantic v2 using schema_to_pydantic_model, type information inside $defs is not preserved. Array types (e.g., {"type": "array", "items": {"type": "string"}}) get converted to empty objects ({"type": "object", "properties": {}}), breaking the schema round-trip.

Reported by: Momo-Not-Emo

The _json_to_pydantic.py utility's schema_to_pydantic_model function recursively converts JSON schemas to Pydantic models. When processing $defs entries, it assumed all definitions were object schemas and created Pydantic models for them, losing type information for non-object types (arrays, unions, literals, etc.).

File: python/packages/autogen-core/src/autogen_core/utils/_json_to_pydantic.py

• Added type-preserving logic for non-object schemas in $defs
• Non-object types (arrays, strings, numbers, booleans, etc.) are now properly handled without unnecessary Pydantic model wrapping
• Added proper handling for $ref resolution to ensure defs are correctly referenced

Medium - The change modifies schema-to-Pydantic conversion logic which is used by structured output features. While it fixes a correctness bug, care was taken to maintain backward compatibility for object schemas. Non-object type definitions that were previously silently corrupted will now produce correct types.