Handle CREATE/CREATED cell processing gracefully under load.
Two issues tangentially related in that both are for handling CPU worker backlog.
We currently sort of have logic for managing the CREATE processing backlog (see:
have_room_for_onionskin), but behavior can be improved (I don't think the aging code actually gets called when we are under extreme backlog since we give up and drop before we age).
With my proposed #13737 (moved) branch, we will now additionally have to contend with CREATED backlog. I will probably opt for similar behavior to CREATED processing (opting to destroy the circuit if the backlog grows too large), but am open to hearing ideas for other options.