Hm… Mounting at root (/) never works, although I don’t know whether that was the sole or even main problem…
(These Fly Postgres clusters really have a lot of moving parts.)
I think I would try the simpler fly m run from the thread above, just so you can poke around a little and verify that this does in fact look like a Postgres volume inside.
(The 528 MB is not reassuring—but may be a mirage caused by the mount conflict.)
If that does look ok, then perhaps the volume forking approach (with an explicit volume ID) will succeed where snapshots failed, 
…