Splitting a row leads to copying all rows below it. During that operation, the information for the row height is lost, when the previous row has a cell spanning on the split one, leading to a missing row in the exported document.
The issue is a regression introduced in R1 2021.