The merge approach you describe is how you would do it. You would just merge the daily files of 10 days into a single merged average. Each day you would create one new 10-day averaged grid based on the grid of the current day and those of the last 9 days. I don’t see anything inefficient in that.