[MLton] free type variables in datatype decs

[Matthew Fluet]

> BTW, Stefan Kahrs wrote "Mistakes and Ambiguities".  Andreas wrote
> "Defects in the Revised Definition" -- and in it notes that the
> restriction on core datatype declarations was removed in going from
> the 1990 to the 1997 Definition.

Agreed on all accounts; except that Andreas titles Appendix A of the 
HaMLet documentation as "Mistakes and Ambiguities", which is what I was 
examining at the time.