Learning Program Representations with a Tree-Structured Transformer