No prod deploy pipeline for the backend — prod-cds runs a locally built image outside CI

[saqibmanan]

prod-cds runs a locally built image (dataexbackend-backend), not the GHCR image the dev pipeline builds and digest-pins. There is no prod deploy workflow at all: deploy-backend.yml and deploy-to-ecs.yml both trigger only on dev and both declare environment: development.

Consequence, right now

Three defects fixed on dev in Sprint 3 are still live on prod, and cannot be shipped by merging (DataSpaceBackend#141):

Verified read-only on prod-cds	State
Conditional user.save() (#136)	absent — row-lock contention present
Health check fresh-connection probe (#135)	absent — cannot detect exhaustion
CUDA libraries removed (#137)	absent — 4.3GB still shipped
uvicorn workers	1

On dev this combination exhausted Postgres connections, which took out the API and the deploy pipeline at the same time, while /health/ kept returning 200 healthy.

Scope

Mirror what DataSpaceFrontend/deploy-Dataspace.yml already does for the frontend — one workflow, branch-conditional environment:

• trigger on main with environment: production
• build once, deploy the digest-pinned image (not a moving tag)
• health gate that can fail (now that /health/ reported the database healthy while Postgres was refusing new connections #140 makes /health/ honest)
• smoke gate via run-smoke.yml@CI with api_base_url pointed at prod
• automatic rollback on smoke failure, run still reporting red
• reachable from main so workflow_dispatch works

Sequencing

The three code fixes should reach prod through this pipeline once it exists, rather than by a one-off manual deploy — that is the point of the epic. If prod needs them sooner than the pipeline lands, do the manual deploy deliberately and note it here.

Prod changes are owner-authorised; nothing has been actioned.