Look into `WITHOUT ROWID` again

Quite a few of our tables use `WITHOUT ROWID`. While this can be an improvement in many situations, it can also lead to bloat if not used with care.

As a short summary, for tables with non-integer PK, an integer rowid is automatically created. `WITHOUT ROWID` prevents this from being done.

This _can_ be expensive, especially if there is no covering index (an index which includes all columns). What occurs is that indices and FK now contain a copy of the PK -- which may or may not be large. If the table has a `rowid`, then the `rowid` is used instead -- which is an integer and therefore smaller and faster to compare and evaluate.

We should look into our `WITHOUT ROWID` usage again, and confirm that its being used appropriately.

This is probably not urgent.