Kubernetes target

Kubernetes is the direction in the handbook and in ADR 0001. It is not how production runs today. Production is Docker Compose on a VPS.

When a move happens, the constraints are already known:

• One deployment per current process, including the relay.
• Probes on /actuator/health (services) and Kong /status.
• Postgres and Redis stay managed (RDS, ElastiCache) or stay as clearly owned stateful sets. Schema-per-service does not change.
• Secrets in the cluster secret store, not in the image.
• At least two replicas for Kong and for the relay.
• The same Prometheus metrics and the same X-Request-Id header.
• A migration cutover that does not run Compose and the cluster against one database as two writers.

Do not add Helm charts "for later" without a decision to leave Compose. A second deploy path that nobody runs will drift from kong.yml.