Ensure failure dlq has same persistence as equivalent in ucam-faas

Description

It is believed that messages entering the "simulated" dead-letter queue (i.e. when a Cloud Run Job fails) aren't persisting (are being cleared automatically/quickly) which doesn't align with normal ucam-faas dlq behaviour.

Potentially due to default behaviour of a pub/sub without a subscription?

Further details

Task list

• Confirm misaligned behaviour (see usage by Google Workspace management functions)
  • Could test in dev area by deploying failing job and checking behaviour
• Add appropriate terraform to align with ucam-faas

Acceptance criteria

• ucam-faas-long's simulated dlq behaves like ucam-faas's real one

Links/references