For the (possibly grouped) resulting relation, the moving average considering width rows is computed where the rows are sorted by the sort_expression list.
When there are fewer than width rows, the average is computed using the current row and all preceding rows.