Redefining Simplicity: Benchmarking Large Language Models from Lexical to Document Simplification