[MLton] caching elaboration

[Matthew Fluet]

> Actually, thinking about it more, I guess I don't care either way about if
> all the errors (parse and type) come for one file before another or not.

I don't really care -- or, more accurately, there are situations where I
want one and other situations where I want the other.  When I've been
porting SML/NJ code, I like the parse errors before type errors, because I
can fairly mindlessly deal with syntax errors and then start thinking
again when the type errors show up.  On the other hand, when touching a
lot of files in the mlton sources, I like to interleave them.