Autoscaling Improvements

Compute autoscaling now reacts faster to traffic spikes, reducing cold-start latency by up to 40%.