Runner-cache builds fall back to public cache when a Kura node becomes unready

When a private Kura (module cache) node was reachable but its /ready probe was failing, builds kept being routed to that dead endpoint and timed out on module cache downloads instead of falling back to the public cache. The runner-cache dispatch is now readiness-aware: it treats the cache endpoint as stale after a 120s window without a heartbeat from the reconciler and falls back to the public cache, so CI and customer macOS runner builds degrade gracefully rather than hang during a cache outage.