restore compatibility with 2.0.0's fix for wraparound bug

[reaperhulk]

fixes #80

[reaperhulk]

This fix is a bit weird to me since arguably OpenBSD's "If you're on the old $2a$ hash you may want this behavior because you're checking existing hashes that were made with that behavior" is more correct, but I guess preserving compatibility with ourselves is more important.

[dstufft]

@reaperhulk Yea-- I think it's especially important to break those particular hashes because they're trivial to brute force (since they hash the same as b"0" * 72) which is one of the first things that something like JackTheRipper will try. However, given that we had the old behavior previously, if we didn't restore it we'd also be breaking anyone who happened to use this kind of password (although the edge case this hits it's quite rare).

Ultimately though, I think the edge case is sufficiently narrow so as not to matter that much, but the fact it generates a warning in passlib motivates me to ensure this is fixed :)

[charmander]

May as well make a positive assertion here?

[reaperhulk]

Done :)