BUG: fixed check _is_unorderable_exception

With upgrade to NumPy 1.13.2 the error message raised when comparing unorderable types
changes from using "'>' not supported between instances of" to using
"'<' not supported between instances of".

This PR checks of these.

See pandas-devGH-17046 for discussion.

This caused failing test test_basic_indexing. Here is the reproducer

```
import pandas
import pytest
import numpy as np
def test_basic_indexing():
    s = pandas.Series(np.random.randn(5), index=['a', 'b', 'a', 'a', 'b'])

    pytest.raises(IndexError, s.__getitem__, 5)
    pytest.raises(IndexError, s.__setitem__, 5, 0)

    pytest.raises(KeyError, s.__getitem__, 'c')

    s = s.sort_index()

    pytest.raises(IndexError, s.__getitem__, 5)
    pytest.raises(IndexError, s.__setitem__, 5, 0) # this part was failing

test_basic_indexing()
```

pandas/core/dtypes/common.py

@@ -1158,7 +1158,9 @@ def _is_unorderable_exception(e):
     """
 
     if PY36:
-        return "'>' not supported between instances of" in str(e)
+        str_e = str(e)
+        return "'>' not supported between instances of" in str_e or \
+               "'<' not supported between instances of" in str_e
 
     elif PY3:
         return 'unorderable' in str(e)