in llm client types add thinking_budget: Option<usize> and set_thinking_budget. don't change any apis. only add set_thinking_budget function.