This just reworks the existing create_irq_on_node() in to the new
create_irq_nr() which is generally exposed. This permits boards that
haven't converted over to sparseirq to try and use their existing ranges,
rather than having arbitrary vectors assigned to them.