Allow returning thought in the tool call responses

[osdemah]

What feature would you like to be added?

Some models include an additional text message in the tool call response. for example:

{
  "role": "assistant",
  "content": [
    {
      "type": "text",
      "text": "<thinking>To answer this question, I will: 1. Use the get_weather tool to get the current weather in San Francisco. 2. Use the get_time tool to get the current time in the America/Los_Angeles timezone, which covers San Francisco, CA.</thinking>"
    },
    {
      "type": "tool_use",
      "id": "toolu_01A09q90qw90lq917835lq9",
      "name": "get_weather",
      "input": {"location": "San Francisco, CA"}
    }
  ]
}

We need to allow including that message in the response, so that it can be properly surfaced to agent and users.

The proposal in #5173 is to change the content field to Union[str, FunctionCalls] instead of Union[str, List[FunctionCall]] so that additional fields could be included in FunctionCalls:

@dataclass
class FunctionCalls:
    function_calls: List[FunctionCall]
    thought: Optional[str] = None

Why is this needed?

to include model response properly

[ekzhu]

Thank you @osdemah , since this is an API breaking change, let's have a bit more discussion around this. We may take a bit longer on this. I do think it is important.

cc @jackgerrits

Is there a way to avoid an API breaking change? So that existing code won't break?

[ekzhu]

@osdemah related #5161 , what do you think about supporting reasoning models together with supporting thoughts in tool calls?