i18n: more accurate line width estimate...possible?

[wimglenn]

For example taking into account full width asian characters, combining accents in latin characters, etc. - it's not entirely trivial because Python module filenames can for example have accents.

This is going to be extremely difficult to handle correctly 100% of the time, without a big performance hit, and this is just a start (plus some testing) for trying to provide a number that may hopefully be more reliable than the string length.

pytest without this stuff:

pytest with this stuff:

[wimglenn]

@nicoddemus any idea what's going on in the appveyor build failure? doesn't look like any module I've touched?

[nicoddemus]

Hmmm seems like a flaky test, let me trigger the build again to confirm