Improve using correct charset after keyword parsing.

After re-joining the lines the text is magically UTF-8, which had been fixed previously by changing the charset of the body to UTF-8 later in the code. This fix clarifies the handling and changes it to use the original encoding, instead of fixing the charset.