When Many-Shot Prompting Fails: An Empirical Study of LLM Code Translation Paper • 2510.16809 • Published Oct 19