0.0.3

fix(vllm): disable thinking mode (chat_template_kwargs enable_thinking=false) and raise max_tokens — fixes empty review on Qwen3 models