Copied from MGMT-13725:

To make the 'none' platform as easy as possible to install, we can make cloud providers use a generic, live disk (for
most if not all of them use virtualization for instances) and leave users with only needing ton download the ignition from the target assisted cluster.

Having a disk, and not a tailored ISO have nice advantages:
1. it is not cluster specific - the disk is just a live RHCOS - could be
distributed more easily, no need for per-provider image
2. it is ready to use for virtualization - it's a qcow disk, easily imported and extended
3. its writeable - the target instance needs just a single disk which will be overwritten by coreos-installer.
4. It makes the number of actions for a user to be up and running smaller when installing more than 1 cluster -
download the disk once, jump start every cluster with just the ignition from the infra-env

To create a live qcow, I did the following:

• downloaded the metal live iso of the version
• convert to qcow using qemu-img
• modified platform from metal to openstack
• modified kernel args to load rootfs to memory
  • initrd to include /images/pxeboot/rootfs.img
  • remove coreos.liveiso=*