An upload API must acknowledge jobs quickly while workers process them asynchronously. Bursts can exceed worker capacity. Jobs must survive worker restarts, and a duplicate delivery must not create a duplicate invoice. Which TWO design choices meet these requirements? (Choose TWO.)

1. Acknowledge every upload before storing the job in any durable service.
2. Place accepted jobs in a durable SQS queue and let workers consume them independently.
3. Store the queue solely in the memory of the current API instance.
4. Use a job identifier and an atomic deduplication check to make invoice creation idempotent.
5. Delete each message as soon as a worker receives it, before invoice creation completes.