[Feature Request]: OpenAI - connect allow_parallel_tool_calls to parallel_tool_calls option

[GICodeWarrior]

Feature Description

The existing allow_parallel_tool_calls option isn't sent to OpenAI. It is used to filter the output in some flows though.

Workaround:
Pass parallel_tool_calls=False via llm_kwargs or similar.

Reason

The lack of this feature can cause issues when expecting a single response (e.g. as_structured_llm). The LLM can split it's response across multiple function calls, and then the framework drops all but the first.

It's made a bit worse since it looks like the framework supports disabling parallel calls via allow_parallel_tool_calls, and that does not currently work.

Value of Feature

No response

[logan-markewich]

I welcome a PR! If not ill probably get to it tomorrow