Some context

So after a careful analysis, there is one case where I am a bit concerned about when not storing null values. AssociationRow.getColumnNames() and Tuple.getColumnNames() would return a subset of columns (only the non null ones).
Investigating now the consequences, it seems there are two problematic cases. In case of query project we have a problem see OGM-564 but we identified this problem already.
The second problem is that when we will implement DIRTY / ALL optimistic locking, we won't consider columns rendered null in the process. That is more concerning.