Calibrating LLMs for Text-to-SQL Parsing by Leveraging Sub-clause Frequencies