Simplify our unpickling logic and add tests for Python 2 pickles

[poodlewars]

This is fixing the issue that Alex pointed out here #2156 (comment) .

We do not need the str case - the stuff we are decoding is always bytes - otherwise we would not be in the pickling fallback. Nor do we need to specify a bytes encoding when decoding Python 2 objects. The "latin-1" fallback case is handled by pd.read_pickle so it's redundant for us to have it too.

The tests I added also pass without my changes to _normalization.py and I think suffice to show there is no behaviour change here.

TODO add tests with a non-standard encoding?

Old Logic

Old logic for pickles inside the msgpack was:

• If pickled in python 2, unpack msgpack with raw=True, pickle.loads with encoding="ascii".
• If pickled in python 3, unpack msgpack with raw=False, pickle.loads with encoding="bytes".

Where raw means:

        """    :param bool raw:
                If true, unpack msgpack raw to Python bytes.
                Otherwise, unpack to Python str by decoding with UTF-8 encoding (default).

        """