A backend API uses an AWS Lambda function behind API Gateway. The first requests after every weekly deployment experience cold starts, causing p95 latency spikes for a few minutes. Which configuration most directly prevents those cold starts for the published version?