chore(runway): run two replicas so a rollout never drains the backend

What

Raise scalability.minInstances from 1 to 2 in .runway/fairway.yaml. Applies to both environments; each currently runs a single pod.

Why

With one replica, every rollout replaces the only backend while the GKE gateway's NEG update lags behind pod replacement. The load balancer serves 5xx and 1.2-1.6s latency that the app never sees: on staging, every deploy window since the SLO registration tripped the RunwayBackendApdexSLOViolation s4 alert (Sep 4's release train and both deploys on Sep 7), while the app's own duration histogram recorded ~1 request over 1s in 8 hours and zero 5xx. Quiet days hold apdex at 1.0. Production runs the same single replica and only avoids the alert by deploying less often.

A second replica keeps one backend serving through the rolling update, so the drain window stops being user-visible.