Prompting Strategy Effects Depend on the Judge: A Multilingual Study of LLM-as-a-Judge Robustness