german umlaut characters break spell checking

[jnweiger]

Words with german umlaut characters e.g. 'Nürnberg' internally seen as "N\xfcrberg" do not make it into the word_set for h.check_word()

This is prevented by the regexp '([a-z_-]{3,})', but leads to encode/decode errors otherwise.
We should try to guess an encoding and if successful convert to utf-8 for hunspell.