Yes, we just didn’t have enough funding/time to implement this as part of the prototype.
Also, there are many ways that simulations deal with varying order. There is nothing in the prototype that prevents using a different cell type only at boundaries as you outlined, or a different approach such as Figure 2 (but applied to the shape attribute rather than fields used for coloring).
For another example, Szabó and Babuska describe CG (not DG) elements which share Legendre shape functions for higher orders on boundaries. In that case, you could have a single connectivity entry for a shared higher-order face, but it would be two numbers (the face ID and an orientation number tagging a symmetric group that matched the rotation+mirror required to put the face into the volume-element’s reference coordinates). So, if you had a quintic face requiring 4x4 = 16 coeffcients for the face’s degrees of freedom, you would only need 2 numbers in the connectivity array for the volume’s face-reference (the start of the 16 numbers plus the symmetric group ID). This is very compact, but also hard to develop+maintain. It also has coefficients that are associated with boundaries of elements but not with specific locations in space (i.e., Legendre != Lagrange). The approach we are proposing would allow this layout as well, but you have to be willing to spend the time to implement it 
.