Use widening to reliably throw InexactErrors

[timholy]

It turns out we still have the behavior described in #21 and #37. Hopefully this PR fixes those issues.

Since this is a change in behavior, I suspect the next tag should be a minor increment (even though @vchuravy and @bjarthur have contributed much more "fundamental" changes, at least their changes were non-breaking). Consequently, now might be a good time to consider other disruptive changes. I therefore also deprecated the lowercase conversion functions ufixed8. I'll probably also work on trying to come up with a more compact printing syntax before we tag.

[timholy]

@JeffBezanson, it looks like you're the only one with "admin" privileges. Can you please turn on the Appveyor webook? I'm hoping to find out whether #15 is still an issue.

[bjarthur]

need to take ufixed12 out of the README

[timholy]

Done. I took the opportunity to expand the amount of explanation about the meaning of these numbers. Had we done this earlier, it might have saved @vchuravy a certain amount of suffering (#32).

[vchuravy]

These changes look great and the additional explanations helps a lot.