Addressing the speed-accuracy simulation trade-off for adaptive spiking neurons