A Toolbox for Surfacing Health Equity Harms and Biases in Large Language Models