With some manual tweaks where futurize doesn't quite do the
right thing.

ReadProxy can not inherit from future object as this breaks the proxy
on Python 2.

The interfaces are compatible so this just changes the namespace.

On Python 3 str has __iter__ so all the tests that are checking
for an iterable in the codebase have to also be careful not to
include str by mistake. Refactor the code that looks for __iter__
to use routines from utils where possible. Where not possible
add instance checks for string.

Use __file__ to find the test file and then derive repository
locations from it.

Future object does not have an __init__ method so we have to catch
that case in the MRO.

Python 3 collections.UserDict is iterable but not on Python 2.
Python 2 requires we use UserDict.IterableUserDict so try that
first and then revert to collections.

This allows the tests to be visible to the test runner and is better
than globally disabling all testing in the file and hiding the tests.

Futurize converted types.StringType to bytes (technically correct
on python2) when it should have converted to str.

The next() method is just a method that iterates but is
not providing an iterable interface allowing the next() function
to work.

pytest assumes that classes with names starting with Test are test
classes that should be analyzed for test methods. When it can't
find any it warns that something might be wrong. Rename the tests
to remove the warnings.