Prompt Injection Attacks on Large Language Models in Oncology