Static scheduling should respect subterm ordering

[robsimmons]

As described in the L10 paper, if we have the following signature

tree: type.
leaf: nat -> tree.
cons: tree -> tree -> tree.

w0: nat -> world.
w1: tree -> world.

w1 (cons X Y) <- w1 X, w1 Y.
w1 (leaf N) <- w0 N.

and 2 X10 places, then when doing static scheduling for the world (cons(cons(leaf 1)(leaf 2))(cons(leaf 3)(leaf 4))), we should be confident that (cons(leaf 1)(leaf 2)), (leaf 1), and (leaf 2) will all be assigned to the same X10 world. This is not what happens in the current static scheduling algorithm, however, as can be seen from running the examples/Back2.l10 example in SMLten.

Place 0, 6 world(s) scheduled
0: w1 (cons (cons (leaf 1) (leaf 2)) (cons (leaf 3) (leaf 4)))
1: w1 (cons (leaf 3) (leaf 4))
2: w1 (leaf 1)
3: w1 (leaf 2)
4: w0 4
5: w0 3

Place 1, 5 world(s) scheduled
0: w1 (cons (leaf 1) (leaf 2))
1: w1 (leaf 4)
2: w1 (leaf 3)
3: w0 2
4: w0 1

(imported from Bitbucket)