fix(litellm): enhance structured output handling

[Arindam200]

Description

Enhanced structured output handling for tool calls in LiteLLMModel.

• Updated response parsing to correctly detect and process tool_calls within LiteLLM responses.
• If no tool_calls are present, the model now parses the message.content directly.
• Added tests to validate behavior for both tool-call and non-tool-call cases.

This ensures structured outputs are consistently parsed across all providers, preventing missing or malformed data in tool responses.

Related Issues

Fixes #1005

Documentation PR

N/A

Type of Change

Bug fix

Testing

How have you tested the change? Verify that the changes do not break functionality or introduce warnings in consuming repositories: agents-docs, agents-tools, agents-cli

• I ran hatch run prepare
• Added and verified unit tests for tool-call and non-tool-call scenarios.

Checklist

• I have read the CONTRIBUTING document
• I have added any necessary tests that prove my fix is effective or my feature works
• I have updated the documentation accordingly
• I have added an appropriate example to the documentation to outline the feature, or no new docs are needed
• My changes generate no new warnings
• Any dependent changes have been merged and published

---

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[dbschmigelski]

I don't think we need this logic, or at least don't need it now.

In the previous state we looked at just choice.message.content even when tool_calls was the finished reason. So the proposal you have here is to switch to look at the function arguments.

I'd rather make one small change at a time, which I believe is simply removing the check as you did in

response.choices[0].finish_reason != "tool_calls"

but I don't think we need to introduce the logic you have here and can instead just rely on the existing as we had before unless there is a compelling reason to

        try:
            # Parse the message content as JSON
            tool_call_data = json.loads(choice.message.content)
            # Instantiate the output model with the parsed data
            return output_model(**tool_call_data)
        except ContextWindowExceededError as e:
            logger.warning("litellm client raised context window overflow in structured_output")
            raise ContextWindowOverflowException(e) from e
        except (json.JSONDecodeError, TypeError, ValueError) as e:
            raise ValueError(f"Failed to parse or load content into model: {e}") from e