backward compatibility for caching done with old pandas indices

Corrects brain-score#11 

I encountered FrozenIndices Error trying to load a pandas dataframe after updating my pandas version. This commit introduces backward compatibility, i.e. you are fine if you pickle.dump() using an old pandas' version and trying to load using a new pandas' version.
pandas-dev/pandas#34535

pickle.load() still called for all non Dataframe objects see https://github.com/pandas-dev/pandas/blob/f2c8480af2f25efdbd803218b9d87980f416563e/pandas/io/pickle.py#L203

Author: p-mc-grath

result_caching/__init__.py

@@ -4,6 +4,7 @@
 import itertools
 import logging
 import numpy as np
+import pandas as pd
 import os
 import pickle
 import xarray as xr
@@ -138,7 +139,7 @@ def load(self, function_identifier):
 
     def load_file(self, path):
         with open(path, 'rb') as f:
-            return pickle.load(f)['data']
+            return pd.read_pickle(f)['data']
 
 
 class _NetcdfStorage(_DiskStorage):