Aligning Cross-lingual Sentence Representations with Dual Momentum Contrast