revise default value for `max_model_len` in LocalLLMServing class

### System Info

N/A

### Others

`max_model_len` should be set to `None` by default, to let VLLM load this value from LLM `config`. Need test to verify the performance.