Abuse through name confusion is possible, of course.
I do not think this can be solved while at the same time supporting i18n.
The real question is: What is the attack vector and how can it be mitigated.
A GNS deployment and the selection of root zones should eventually converge towards a trusted namespace (for each individual user).
There is not really any incentive having a TLD zone in your configuration that would allow such odd labels as there is not much use beyond fraud in using them.
The governance policies ("rules") of the zone are expected to "trickle down" the delegation chains or alternatively damage reputation of the parent zone.
The labels in a name are separated using the character "." (dot).
Any mention of what happens with other dot-like codepoints?