Tables in PDFs very often have no lines around them at all. Where lines do exist they are drawn on a different layer from the text, invisible to anything pulling the words out. That is why approaches built on finding the lines fail so often.
So position is what gets read. Within a line, a gap much wider than a word space means a cell boundary, and when those gaps stand at the same horizontal position across several lines in a row, that block is a table. Three lines is the minimum. Two-line blocks are usually a title with a date pushed to the far right, and turning that into a table makes it harder to read, not easier.
An empty cell leaves no trace in the file, because there were never any characters there. Joining cells up in order would shift everything along by one, so each cell is dropped into the column nearest to it and the rest are left blank.