Remove StreamTagger and refactor the code

[hkazuakey]

I think StreamTagger can be removed since it is not refereed from any other codes. At the same time, several legacy codes can refactor to modernize the code. For example, Gosen should utilize char.def which is an external configuration file that can specify how to handle characters by its type, which is hard-corded in Gosen at this moment, and that enable us to freely modify the tokenization behavior.