An application uploads large S3 objects over unreliable links. Retrying an entire object is expensive, and abandoned attempts accumulate storage charges. Which approach best addresses both concerns?

1. Use multipart upload with per-part retries and an appropriate abort-incomplete-upload cleanup policy.
2. Retry all successful parts on every failure and retain incomplete uploads forever.
3. Omit CompleteMultipartUpload and treat the upload ID as the finished object.
4. Use only S3 Versioning and assume it resumes partially uploaded bytes automatically.