Towards Trustworthy and Aligned Machine Learning: A Data-centric Survey with Causality Perspectives