fix(core): preserve $defs array/primitive types in schema_to_pydantic_model

[Mr-Neutr0n]

Summary

Fixes #7203

When $defs contains non-object type definitions (arrays, primitives), the schema_to_pydantic_model function was incorrectly creating empty BaseModel classes instead of preserving the original type information.

Before (broken):

# Original schema
{"$defs": {"TaskData": {"type": "array", "items": {"type": "string"}}}}

# After conversion - TaskData becomes empty object ❌
{"$defs": {"TaskData": {"properties": {}, "title": "TaskData", "type": "object"}}}

After (fixed):

# TaskData is correctly preserved as array type ✅
instance = Model(task_data=["item1", "item2"])  # Works!

Changes

• Add _type_cache dictionary to store non-object type definitions separately from model cache
• Add _schema_to_python_type() method to convert non-object schemas (arrays, primitives) to Python types
• Update _process_definitions() to detect non-object types and route them appropriately
• Update get_ref() to check type cache before model cache

Test Plan

• Added test test_defs_array_type_preserved - reproduces exact issue from schema_to_pydantic_model loses $defs type information (array → object) #7203
• Added test test_defs_primitive_type_preserved - tests primitive types in $defs
• Added test test_defs_array_with_constraints - tests array constraints (minItems/maxItems)

[Mr-Neutr0n]

@microsoft-github-policy-service agree