[2403.12025] A Toolbox for Surfacing Health Equity Harms and Biases in Large Language Models