Copy propagation on `Get` needs a small modification

[minhddo]

The current implementation of copy propagation at word_copy doesn't optimize away multiple Gets from the same stores before a Set as shown in this example

      (17 <- HeapLength)
      (inst (Arith (Shift Lsl 21 17 1)))
      (25 := CurrHeap Add 21)
      (inst (Mem Load 29 (Addr 25 0xFFFFFFFFFFFFFFF8)))
      (inst (Arith (Binop Add 33 29 (Imm 0x1))))
      (37 <- HeapLength)
      (inst (Arith (Shift Lsl 41 37 1)))
      (45 := CurrHeap Add 41)
      (inst (Mem Load 49 (Addr 45 0xFFFFFFFFFFFFFFF8)))
      (inst (Arith (Shift Lsl 53 49 2)))
      (57 <- HeapLength)

The current implementation allows multiple loads from HeapLength to persist, while these can be changed to Move instead.