It's a nice way of saying that there is still one failing, consistently.

Bytes should always be returned from the streams, so the tests
should be checking against byte strings instead of text strings.

This also fixes the `sha_iter` as it relied on the Python 2
`iterkeys` which has been renamed to `keys` in Python 3.

There were a few things which were being reused consistently for
compatibility purposes, such as the `buffer`/`memoryview` functions
as well as the `izip` method which needed to be aliased for
Python 3.  The `buffer` function was taken from `smmap` [1] and
reworked slightly to handle the optional third parameter.

This also adds a compatibility file dedicated entirely to encoding
issues, which seem to be the biggest problem.  The main functions
were taken in part from the Django project [2] and rewritten
slightly because our needs are a bit more narrow.

A constants file has been added to consistently handle the
constants which are required for the gitdb project in the core
and the tests.  This is part of a greater plan to reorganize
the `util.py` file included in this project.

This points the async extension back at the original repository
and points it to the latest commit.

[1]: https://github.com/Byron/smmap/blob/1af4b42a2354acbb53c7956d647655922658fd80/smmap/util.py#L20-L26
[2]: https://github.com/django/django/blob/b8d255071ead897cf68120cd2fae7c91326ca2cc/django/utils/encoding.py

This fixes most of the import errors that came from using the
implicit relative imports that Python 2 supports.  This also fixes
the use of `xrange`, which has replaced `range` in Python 3.  The
same has happened for `izip`, which is also being aliased.

The octal number syntax changed in Python 3, so we are now
converting from strings using the `int` built-in function, which
will produce the same output across both versions of Python.