Modify tests for text wrongly split into encoded words

We were using test data submitted by users. But we don't really know
what the expected result should be. In the second test the expected text
ended in a replacement character. That indicates the text is not decoded
correctly. Maybe there was an additional encoded word that was missing.