Append agent.say after function_call output

[longcw]

After test #1130 for some cases, I found it better to append the contents of the say called in the function after the tool_output so that LLM could complete the full answer exactly following the said content.

Can we add such a flexibility for user to set a extra_chat_messages in AgentCallContext that will be appended after the extra_tools_messages?