fix: Fix the issue where some model providers do not yet support max_completion_tokens, causing the inability to set the maximum number of response tokens.#7592