Apply Unicode Normalization

[Marcono1234]

What do you think about applying Unicode Normalization (German Wikipedia article) to both language model creation and language detection?
This would allow accurate language detection even if input text or training data is non-normalized. Java provides the Normalizer class for this, see also Java Tutorial. For Lingua one of the canonical composition forms (NFC or NFKC) should probably be used (with the decomposition forms Lingua's checks for unique letters would not work anymore). NFKC might be best since it also avoids some redundancies (though it could in some rare corner cases also introduce inaccuracies).

[Marcono1234]

@pemistahl, what is your opinion on this? I assume it could help increase the accuracy when text uses separate diacritics characters.