Roll the API fleet without losing capacity

A previous refresh of the production API fleet replaced too much capacity at once and interrupted requests. Launch template version 7 has now passed testing, and the api-refresh-errors alarm identifies a bad rollout. The replacement must preserve serving capacity and avoid recycling instances already on the approved configuration.

Repair refresh-fleet.sh to roll api-production to tested launch template lt-0a13d9c4e8f7246bb version 7.

• Preserve 100% healthy desired capacity with at most 150% temporary capacity, skip matching instances, and roll back automatically if api-refresh-errors enters ALARM.
• Keep the Rolling strategy and fail safely on script errors.