In the patch "golang_str: bstr/ustr index access" we added __getitem__
implementation for bstr/ustr and thorough corresponding tests to cover
all access cases: [i], [i:j] and [i:j:k].

The tests, however, are run via pytest which does AST rewriting, and, as
it turned out, always invokes __getitem__ even for [i:j] case even on py2.
Which differs from plain python2 behaviour to invoke __getslice__ for
[i:j] case if __getslice__ slot is present.

Since on py2 both str and unicode provide __getslice__ implementation,
and bstr/ustr inherit from those types, they also inherit __getslice__.
And oops, then on py2 e.g. bstr[i:j] was returning str instead of bstr:

    In [1]: bs = b('αβγ')

    In [2]: bs
    Out[2]: b('αβγ')

    In [3]: bs[0]
    Out[3]: b(b'\xce')

    In [4]: bs[0:1]
    Out[4]: '\xce'              <-- NOTE not b(...)

    In [5]: type(_)
    Out[5]: str                 <-- NOTE not bstr

-> Fix it by explicitly whiting out __getslice__ slot for bstr and ustr.