DETAIL Matters: Measuring the Impact of Prompt Specificity on Reasoning in Large Language Models