The NormalizationTransliterator (icu/source/i18n/nortrans.cpp) uses UnicodeSets
of "skippables" to find out which code points are safe to include in a buffer,
i.e., which do not interact with any further code points.
It is used to normalize as much of a buffer as possible when the buffer is not
at the end of the input stream.
The "skippables" data should be computed by gennorm and stored in unorm.dat.