[MLton] changes to compile with -sequence-unit true

[Matthew Fluet]

> Matthew, your recent checkin fails to compile with SML/NJ.  The
> problem is SML/NJ's weaker handling of the value restriction (see
> directed-graph.sml).  In general, there are also potentially problems
> with SML/NJ's weaker inference of flexible record types.

Sorry.  I'll keep that in mind.

> I guess I'll take this as good evidence that you work pretty much
> exclusively with the new front end :-).

Yup.  MLton dumps an error message well before SML/NJ.