Batch payload processes retry object store uploads

Large batch processes are no longer derailed by momentary network hiccups. Previously, a single "fetch failed" error when communicating with the object store would immediately abort an entire stream of batch items.

The batch payload processor now automatically retries object store uploads up to three times under the hood, using an exponential backoff. If the server completely exhausts its retry attempts, the error response now permits the client SDK to trigger its own fallback recovery sequence. This makes the batch processing pipeline significantly more resilient, ensuring that transient connectivity drops self-heal without interrupting active workloads.