Traverses a computation to reach the values within. Any values of the form

  both a (both b c)

are transformed to

  map3 a b c ~f:(fun a b c -> a, (b, c))

since this generates fewer incremental nodes, thus leading to better performance. The same optimization gets applied for up to 7 nodes both'd together. A similar optimization is applied to nodes of the form

  map (both a b) ~f

which get turned into

  map2 a b ~f:(fun a b -> f (a, b))

since, again, this generates fewer incremental nodes. As with the previous transformation, this generalizes to an arity of 7.