Write canonicalizer for Racket core dataset

[rrnewton]

This can be a simple racket program that reads in the current batch of expaded data and produces something in a reduced grammar. This reduced grammar will have a direct interpretation as an algebraic type in our language.

Proposal from slack:

(1) I propose that we emphasize "expr" and squish away all top-level / module-related forms to the following:

(data prog [prog (list toplvl)])
(data toplvl
      [define-values (list Sym) expr]
      [define-syntaxes (list Sym) expr]
      [expression expr])
(data expr [VARREF Sym] [plain-lambda ...]...)
(data formals
      [F1 (list Sym)]
      [F2 (list Sym) Sym]
      [F3 Sym])

(2) we keep all the constructors of "expr" in the same order as the Racket docs, and that determines their tags, starting with 0

(3) for our own purposes we introduce the tags VARREF, F1, F2, F3 above, which are not part of the original grammar

(4) we build-in a notion of list fields in our ASTs, e.g. (list Sym). Worst-case they could be new, packed cons-list datatypes, but we can do better if we bake them in.

(5) by convention we strip all occurrences of #% prefixes in tags

[rrnewton]

1st shot at defining a standard grammar for the canonicalized rep -- 775b946

[samth]

I slightly modified the grammar, and implemented this in rewrite.rkt.