Update CONTRIBUTING and HACKING to reflect policy on Python 3 contributions.

[jaraco]

Consider updating the documentation to dissuade contribution early rather than cause surprise later.

[sivel]

This is not true. To provide additional information around python 3 and the PR that was recently submitted, it fixed 1 python 3 incompatibility to uncover another.

The preference would be to perform larger scale fixes. If a single targeted fix is applied, especially to a lower level function, it could cause other higher level functions to not function properly.

It would be my recommendation to ensure that the code in the unit tests are python 3 compatible, and then validate that the changes to the tests do not interfere with the code being tested. At which point, larger scale changes that target all python 3 incompatibilities can be applied. Such python 3 incompatibility updates should at first not change any functionality. The same exceptions raised currently, should be raised after the updates. Later, more invasive changes could be applied once we have a good baseline of the app functioning properly in both python 2.7 and 3.3/3.4.

To get something like clouddns.py to properly support python 3 will require some changes to http.py which will then potentially affect everything else in the project.

Additionally, I have recently just become a maintainer of this project, and I do not have a high level of confidence that changes like this can be made safely with the unit tests as they currently exist. I would feel uncomfortable potentially damaging functionality that works in python 2.7 to add python 3 support.

[jaraco]

Well, that's good. It just seemed like copy/paste response to three Python 3 tickets was indicative of a shift. That helps clarify.

I've ported scores of packages to Python 3, so I'm very familiar with the process. Unfortunately, I don't have time to dedicate to porting the whole library, so I'd like to bite off small pieces as time permits.

Based on your feedback, a patch like #477 should cast to a list in order to be very conservative about any behaviors, including unexpected exceptions. That seems reasonable to me, and why I mentioned it in the PR.