[MLton] HOL's Moscow ML implementation, and pushing MLton to emulate it

[Stephen Weeks]

> Here's one example of how to deal with polymorphic recursive
> structures in a uniform way without recursively coercing the whole
> structure into a univerisal rep.

Makes sense, but I'm not sure that it helps with the tabulate1 problem
I mentioned, where one has no choice but to produce the entire
specialized representation to pass to the host function.  You might
also see if you can apply laziness tricks to solve the map example in
Nick's paper, since he's clearly thought about it and thinks laziness
doesn't help.