Recover checkout endpoint selection

Repair checkout.yaml: exactly checkout Deployment (apps/v1) and Service (v1), both explicitly in shop. Preserve two replicas, checkout container, registry.example/checkout:4 image, checkout-runtime service account, and template labels app=checkout,component=api,release=current; extra labels are allowed. The nonempty Deployment selector must match its template (matchLabels or matchExpressions). This reconstructed source is evaluated before application; changing an existing live Deployment selector requires replacement planning. Keep one internal, non-headless ClusterIP TCP port 80 targeting actual listener 8080, numerically or through a valid named container port. Select every intended Pod, including warming Pods, and no unrelated Pod. Route only to all ready intended Pods; do not publish unready addresses. Intended Pods inherit template labels plus observed track/state labels. Steady: api-a/api-b ready,track=stable,state=ready; api-warm unready,track=canary,state=warm. Replacement: api-b ready/stable, api-c ready/canary, api-warm2 unready/stable, with the same ready/warm state labels. Unrelated shop Pods: old-api (app=checkout,component=api,release=old,track=stable; ready; http=8090), diagnostic (app=checkout,component=diagnostic,release=current; unready; http=9000). Ready foreign-api in sandbox has the supplied identity labels and http=8080. Both stages must work. Listener ports and readiness are immutable observations. The model supports the supplied fields and selectors, not other Pod features or controller, probe, policy or packet execution. Inspect selected/eligible identities after each repair and recover without Reset.