Push the hello-world array literal example all the way through

[rrnewton]

Ok, so this is something that's happened in bits and pieces in different places. @pavenvivek's ffi-test is ALREADY calling full Accelerate map and zipWith functions, in monomorphic varieties.

What is missing is just an end-to-end demonstration that:

• starts with the end-user Racket syntax
• goes through the front-end "compiler", even if it just a stub (see #8)
• goes through an SExp + C-data form to get into Haskell
• is processed by Haskell (existential types, fanciness, call Interpreted.run)
• is returned to Racket, reconstructing a valid acc-array?

Did I miss anything? Let's close this issue when we have such a thing, and discuss any problems here.

@kylcarte, what is the minimum we need from Marshal.hs or something similar to get this hello world working?

[rrnewton]

This works with @pavenvivek's prototype.

I think there's some work left to do for us to run these examples in the manner we ultimately mean to be running them. I.e. with a correct compiler pipeline, staging, data representations -- incomplete only in "width" (unhandled cases) not depth or kind.

[rrnewton]

Ok, we are getting close to this in terms of the front-end. The simplest programs will already be:

1. Fully type annotated
2. Fully normalized

So typecheck and normalize will just be NOOP on these simple programs. The remaining blocker then is proper SExp parsing on the Haskell side, Hint-call, and return results.