Beyond Accuracy: Evaluating and Explaining the Capability Boundaries of Large Language Models in Syntax-Preserving Code Translation
