Cloud provisioning to tenant inference in one flow.

Staff create tenants, assign physical devices, push base config, then customers manage telemetry, model installs, vLLM deployments, and local API keys from their portal.