Currently, VirtualBMC works by forking a new process on every 'vbmc start' invocation. This makes it hard to manage it by standard means, including starting it on system boot and restarting when it fails.

Instead, we need one daemon (vbmcd?) serving IPMI for all virtual nodes, and the vbmc command merely talking to it. Of course, we'll need to somehow maintain a reasonable upgrade path for both upstream CI and TripleO use cases.

We'll have to update packaging to include systemd units. And we'll probably need puppet-ironic and instack-undercloud code managing the new service (now it's done either manually or via tripleo-quickstart).