Avoid duplicate calculation within a single rule

[robsimmons]

This is an issue that primarily comes up if you write

#builtin NAT_SUCC s
a :- b X, c (s (s (s X))), c (s (s (s (s X)))).

This gets compiled to

a :-
  b X,
  s X is X1
  s X1 is X2,
  s X2 is X3,
  c X3,
  s X is X4,
  s X4 is X5,
  s X5 is X6,
  s X6 is X7,
  c X6.

which should be able to be simplified to

a :-
  b X,
  s X is X1
  s X1 is X2,
  s X2 is X3,
  c X3,
  s X3 is X7,
  c X7.

This is basically common subexpression elimination, and it should (probably) be performed during compilation on the flattened program, before binarization.