Is this being done because it can be, or is it solving a real-life pain point? Just wondering what the perspective is here.
Much of partitioning strategy seems to me to revolve around how the system is used, and not just the schema and what is possible. For instance, you can mimic primary and foreign key behavior with triggers as described here, and that would bypass some of the restrictions on what can be done.
This would allow you to change out the primary key for a simple index perhaps, and partition however you want. Just because something can be done, does not mean it should be.