set up oVirt/RHV in a perf lab for OCP testing

There are many assumptions based on Red Hat's Perf&Scale labs

Design

oVirt setup with VMs running OCP cluster using isolated networks

1 physical host installed with oVirt Engine, hosting dnsmasq for OCP network, routing to external networks, hosting NFS for VM initialization
n-1 physical hosts installed as oVirt hosts, using vdsm-hook-localdisk for local storage. Storage is initialized when VM starts on a host for the first time, then it's reused. VMs are also pinned on their first power on via a vdsm-hook call back to API.
OCP "baremetal" network running dnsmasq, routed through the engine host to public
OCP "provisioning" network for PXE baremetal provisioning

{
    "nodes": [
        {
            "pm_addr": "mgmt-FQDN",
        }
    ]
}

1 external interface for connectivity, 2 interfaces connected to all hosts without DHCP, for OCP's baremetal and provisioning networks.
interfaces need to be mapped from the inventory host names. Adjust in 05-get-inventory.sh or just create the right mapping yourself by creating files engine and hosts with

 <fqdn>,<external_iface>,<baremetal_iface>,<provisioning_iface>

can be run individually for troubleshooting/reinstalls

process inventory json and extract the first node as the engine host and the rest as oVirt hosts
setup ssh keys (executor->engine+hosts, new engine key, engine->hosts)
upgrade to 8.6, install packages
insert reboot
set up engine hosts's networking and install oVirt engine
set up local storage on hosts
add oVirt hosts
add master and worker templates
create all individual VMs
start vBMC servers

starting VMs. Not needed for IPI so maybe not...
better ovirtvbmc control. oVirt/vbmc#3
skipping masters to support baremetal masters + virtual workers
configure prometheus to run on masters, by default it runs on a random worker node and it becomes a huge memory hog with >100 nodes
local mirror of RHCOS and/or OCP registry mirror