Text this: TableStructureFormer: an improved masked-attention mask transformer model with long-distance feature aggregation and deep detail supervision for table structure recognition