normalize lines before getting charset

[mshannon-sil]

In find_missing_characters(), we are currently normalizing after gathering the set of all characters, but we should instead be normalizing text before using it to update the set of all characters. Same issue as sillsdev/silnlp#352