Understanding and Meeting Practitioner Needs When Measuring Representational Harms Caused by LLM-Based Systems