Nested Pydantic models stop being recognized after one inheritance layer

[YizukiAme]

Describe the bug

In deepeval/prompt/utils.py, the structured output schema builder detects nested Pydantic models with BaseModel in annotation.__bases__. This only works when the field type directly subclasses BaseModel. If a project uses an intermediate base class (e.g. class DerivedPayload(BasePayload) where BasePayload(BaseModel)), the check fails and the field falls back to STRING type.

To Reproduce

from pydantic import BaseModel

class BasePayload(BaseModel):
    value: int

class DerivedPayload(BasePayload):
    extra: str

class MyOutput(BaseModel):
    payload: DerivedPayload  # This gets typed as STRING instead of OBJECT

# Because:
print(BaseModel in DerivedPayload.__bases__)  # False
print(BaseModel in DerivedPayload.__mro__)    # True

Root Cause

deepeval/prompt/utils.py L237-252:

elif (
    hasattr(annotation, "__bases__")
    and BaseModel in annotation.__bases__
):
    field_type = "OBJECT"

Expected behavior

Any Pydantic BaseModel subclass (including indirect inheritance) should be recognized as an OBJECT type in generated schemas.

Suggested Fix

Use issubclass(annotation, BaseModel) or check __mro__ instead of __bases__, consistent with how array item models are already handled elsewhere in the same file:

elif (
    hasattr(annotation, "__mro__")
    and issubclass(annotation, BaseModel)
):
    field_type = "OBJECT"

[SamSi0322]

Taking this together with #2609 in a single PR — both bugs are in the same function.