“Dr LLM, what do I have?”: The Impact of User Beliefsand Prompt Formulation on Health Diagnoses