Practical Membership Inference Attacks against Fine-tuned Large Language Models via Self-prompt Calibration