While I agree with not trying to reconnect on shutdown, that predictable fail due to absence of iscsid, I am not sure what is the motivation of other changes. For example, destroying device before the final sync may leave FS dirty. Is there some problem to keep connection active until both FS and CAM do their syncs?

Erm, but that's exactly what this patch does - or at least it's supposed to do: working sessions continue to work as before, while disconnected ones fail IO instead of waiting for reconnection.