SlangDIT: Benchmarking LLMs in Interpretative Slang Translation