Using an answer instead of a comment because of limited space in a comment.
Your document uses 2-column sections. Within these section you insert a 2-column table. This creates a complicated context.
Internally, the table is built as if the section had an infinite length. This gives the full height of the table. The table is then split in order to try to balance height in section columns.
Your section 1 exhibit this behaviour with a split at line closest to mid-height. Notice that portions of the table are not equal in their respective columns.
In section 2, you have a shorter table and the algorithm splits the table closest to mid-height too. Once this is done, text is laid out, without consideration for orphan/widow because split decision is already made.
I think there is a complex interaction between two split criteria: the one in paragraph text flow and the one in table text flow. In my opinion, the table criterion takes precedence over the paragraph because the table is the “outermost object” to be formatted in the page. The paragraph is contained therein and is laid out within the limits of the table and its cells.
Section 3 is a bit different. Space remaining at the bottom of the page would be sufficient if the table were split exactly at its middle. However since lines must be shown in their entirety, first column is cut shorter than the middle, which leaves a taller second column. This second column meets the bottom of the page. Since split decision is already made, there is no reason to split again (for a few pixels only) otherwise you probably enter an infinite decision loop. Because margins are a strict no-print zone, text is clipped and descenders disappear.
Note: I met such a clipping with frames extending too far in their container.
Weird behaviour: if I play with the section 2 table properties (split and break attributes), the table displays either as single rows in separate columns or full table in first column. Restoring the attributes does not revert to initial formatting (even with Tools
>Update
>Page Formatting
) ! I must add content in a cell and remove it to cause a real page formatting update.
I noticed this strange behaviour in a complex document of mine with same characteristics: 6-8 columns section with a 3-column table. Editing this table is a deep nightmare with sometimes full columns becoming empty (content flushed to next column), spurious page breaks or uneven columns. The cure is always add-then-remove content. However, this will not solve the widow/orphan issue.