Why does high-frequency quantitative trading emphasize low latency so much? What is the significance of continuing to optimize to the microsecond level after reaching the millisecond level?