Added two predictions:
  GET_ITER --> FOR_ITER
  FOR_ITER --> STORE_FAST or UNPACK_SEQUENCE

Improves timings on pybench and timeit.py. Pystone results are neutral.