Make the checkout fleet come back down

Rewrite autoscale.json so the checkout fleet grows under sustained load, comes back down under sustained quiet, and ignores a short spike. Keep the setting and every rule pointed at the supplied scale set and the supplied metric. Declare a capacity minimum, maximum and default inside the approved fleet range. The starter has only a scale-out rule, which is why the fleet reached its ceiling in the peak and stayed there, so add the matching scale-in rule. Then choose the two thresholds together: they must be far enough apart that the capacity a scale-out reaches does not immediately satisfy the scale-in rule, or the loop reverses on the next evaluation. Finally make the scale-out look-back longer than the stated four-minute spike so a blip does not add instances.