Transformers with Sparse Attention for Granger Causality