There is usually a good case for retaining some physical server(s); or forgoing a cloud journey altogether. In this case, what is a good strategy for ongoing management of physicals?

Well - it’s kinda the same strategy for ongoing management of Virtual/Cloud …

There have been excellent improvements in package management over the years; and now it is quite possible to do in situ OS upgrades - minimal downtime.

With just a single server (or a small number in your cluster), it is probably not worth the effort to deploy OpenStack or Kubernetes; it it still worth using virtualisation or containers - just not with a heavy set-up/management overhead.

We use libvirt/qemu for to do this; the physical host is deployed with minimum OS; libvirt and a load balancer to ship traffic to appropriate VM to service the request.