恐怕带有PARTITION BY user_id子句的窗口函数row_number()与您的情况无关，因为它只关注当前行的user_id，而您想要的是包含所有用户的合并视图。我没有比在表sales上执行自连接更好的主意了：

WITH list AS (
SELECT DISTINCT ON (s2.date, user_id)
       s2.date
     , product
  FROM  sales AS s1
 INNER JOIN (SELECT DISTINCT date FROM sales) AS s2
    ON s1.date <= s2.date
 ORDER BY s2.date, user_id, s1.date DESC
)
SELECT date, product, count(*)
  FROM list
 GROUP BY date, product
 ORDER BY date

请参见dbfiddle中的测试结果