Fix bare-metal Kura instances stuck behind an unschedulable dedicated gateway

On bare-metal (host-network) Kura regions, accounts previously flagged for a dedicated gateway would get stuck unschedulable because the per-account host-network gateway collided on port 443 with the shared platform gateway. Dedicated accounts on host-network regions now fall through to the region’s shared ingress class served by the platform gateway, so Kura instances actually serve traffic. LoadBalancer regions are unchanged and still get a dedicated gateway per account. Instances already provisioned on a host-network region keep their old class until the next re-roll; rerunning a Kura instance via the dashboard’s Destroy then Deploy reattaches it to the shared class and cleans up the orphaned dedicated gateway.