Operate on characters rather than bytes in the dialog so that it
actually works with Kanji.
Rewrite the auto-matcher to handle more cases and add unit tests for it.
This is currently only correct if the UI language is the same as the
subtitles language as the global locale is used, but it should still
never be worse than the small hardcoded table of word splitters.
Closes#1206.
Do proper unicode case-folding for case-insensitive searching rather
than converting only ascii characters to lowercase. The Turkish 'i' is
still not handled correctly (since it's the only place where
case-folding is locale-dependent), but that's probably not worth caring
about as long as we don't have a Turkish UI translation.
This affects both the find/replace dialog and the select lines dialog.
Closes#1342.