It's a requirement of the uncertainty principle/noncommuting observables that some of the uncertainty in quantum theory (without hidden variables) is ontic
I think this statement, like your previous one about density matrices, is interpretation dependent. As I understand the TI, the inability to simultaneously make exact measurements of non-commuting observables is due to the dynamics; it's not due to any ontic uncertainty in the state. Unless that counts as "some of the uncertainty is ontic", then "some of the uncertainty is ontic" would seem to me to be interpretation dependent.
Sometimes a subsystem is not in *any* local eigenstate
The TI doesn't say eigenstates are ontic, so this is irrelevant to the TI. The TI says q-expectations are ontic.
An example is one of the qubits when the state is a Bell state. In this case, we say the qubit's state is an improper mixture, rather than a pure state.
But each qubit still has q-expectations, and those aren't uncertain.
If the uncertainty that necessitated the use of an EV was merely epistemic, just a matter of our ignorance about the state, it would make no sense to ever suggest the EV was a beable.
You've got the TI backwards. If the q-expectation is a beable, then what doesn't make sense is to talk about "the uncertainty that necessitated the use of an EV". The EV is the ontic state; any talk about other objects like state vectors or density matrices is what would be necessitated by uncertainty, i.e., our inability to make infinitely precise measurements of the q-expectation, the ontic state.