In the next patch we will start comparing dumped pickles for bstr/ustr
to expected golden data. However there are many irrelevant differences
in how different pickle modules, and protocols produce them:

- cPickle and pickle.py differ from where they start to count *PUT
  indices. cPickle start to count them from 1, while pickle.py from 0:

  In [1]: s = '123'

  In [2]: cPickle.dumps(s)
  Out[2]: "S'123'\np1\n."

  In [3]: pickle.dumps(s)
  Out[3]: "S'123'\np0\n."

- sometimes unused *PUT and MEMOIZE opcodes are emitted.
  see previous item for an example.

We will filter out those details and compare to golden data the
resulting pickles after normalization, which brings pickle data into
normalized form, which still, when loaded, will be loaded in to the same
object as original pickle, but where there are no unused *PUT/MEMOIZE
opcodes and where *PUT indices always start from 0.

-> Add pickle_normalize utility as a preparatory step for that.