unicode-width should retain a semi-stable width function

As described in #55, 0.1.13 changed the predicted widths of a lot of characters to be more accurate for things like emoji.

From that issue, and from the breakage in rustfmt (rust-lang/rustfmt#6203), it feels useful to have an API that is explicitly stable, and documented as such. Probably one that did the original less-useful-but-"still works" thing of relying on East Asian Width properties only.