Enhancing Trust in LLM-Generated Code Summaries with Calibrated Confidence Scores