For a arbitrary Apter Tree, a linear scan is unordered. You can impose additional constraints to get a ordered traversal (in the same way that, eg, you can sort a assoc-list/JSON-style key-value table by keys to get a key-order traversal), and the result is still a valid Apter Tree (respectively valid list of key-value pairs).
Yes but that is not what is presented (a B+ tree is not a B tree even with minor modifications) and it changes the complexity of your other update operations drastically. The thing that grates me (as someone that has written a dozen or so different tree structures) is that this one is presented as a particularly good one, and I think it excels at almost nothing, hence its obscurity.