> If some duplicate inputs would have been permitted without crashing nodes then the update appears to fit the definition of soft fork.

If this were the case it would be a complete failure of the system, rather than a DoS.