Revisit dead letter persistence for delivery failures

[stadia]

Summary

We intentionally removed dead letter record persistence and related tracking from feature/delivery-reliability because it is not needed for the current delivery scope.

Follow-up

Evaluate whether we still need a persistent record for failed deliveries in a later iteration.

Potential scope:

• dead letter model/table for permanently failed deliveries
• operator-facing tracking and retry workflow
• cleanup or retention policy for stored failures
• criteria for when permanent failures should be persisted vs simply discarded

Context

Current behavior keeps retry handling for TemporaryDeliveryError and discards PermanentDeliveryError without storing extra state.

This issue is a reminder to revisit the product and operational need before reintroducing persistence.