Response cache tells same-named output schemas apart

Agno's Model._get_model_cache_key() now turns a Pydantic response_format class into its model_json_schema(), nested field definitions included. Earlier releases used the class name, so two Output classes in different modules collided whenever the prompt matched.

With cache_response=True, the second agent got the first agent's cached JSON. We gave two agents one OpenAIChat with cache_response=True, pointed it at a mocked endpoint that answers whichever schema it receives, and asked both the same question. One output_schema was Output with a total: float field, the other Output with priority: str.

# Before (real output)
cache keys equal: True
Output(total): Output(total=42.0)
WARNING Failed to parse cleaned JSON: 1 validation error for Output
priority
  Field required [type=missing, input_value={'total': 42.0}, input_type=dict]
WARNING Failed to convert response to output_schema
Output(priority): '{"total": 42.0}'
requests sent to the provider: 1
 
# Now (real output)
cache keys equal: False
Output(total): Output(total=42.0)
Output(priority): Output(priority='high')
requests sent to the provider: 2

Classes with identical schemas still share an entry. A model with no JSON schema, such as one with a Callable field, falls back to its module and qualified name.

See the cookbook, and learn more about response caching in the documentation.