A Lambda consumer reads SQS batches. One record fails after the others succeed, and downstream writes must not duplicate during retries. Which implementation choices are appropriate? Select 2 answers.

1. Delete all messages immediately before processing the batch.
2. Return success for every failed record without preserving it.
3. Make target writes idempotent and configure a visibility timeout suitable for processing and retries.
4. Enable supported partial-batch responses and identify failed messages in the required response shape.
5. Assume standard SQS provides exactly-once application side effects.