Assistant is thrown into infinite loop if assistant.tool_choice = "function_name" is selected and autorun.

[andreibondarev]

Describe the bug
The Assistant is thrown into an infinite loop when LLM is forced to pick the same tool on every request. Since we auto-run the tool calls and then call the LLM again -- it's thrown into an infinite loop.

To Reproduce

llm = Langchain::LLM::OpenAI.new(api_key: ENV["OPENAI_API_KEY"])
assistant = Langchain::Assistant.new(
  llm: llm,
  instructions: "You're a helpful AI assistant",
  tools: [Langchain::Tool::NewsRetriever.new(api_key: ENV["NEWS_API_KEY"])],
  tool_choice: "..."
)

assistant.add_message_and_run!(content: "What's the latest news about AI?")

Observe the Assistant thrown into an infinite loop because:

1. It calls the LLM which is forced to return a tool call
2. Tool call is auto-run
3. It calls the LLM again which is forced to return a tool call again
4. And it keeps repeating

Expected behavior
The openai/swarm framework allows setting max_turns: which (I believe?) stops this from happening. We may consider something similar.