TODO: if the non-fitting word was made up of components from more than one alphabet (roman, italic, etc), the word will be removed and placed back on the line using only one alphabet, namely the one containing the last part of the word that fits. That's because we don't associate the alphabet with the text to be placed on the line as part of the setting buffer.
We should probably change the setting buffer to contain the alphabet (and other settings) of the word, instead of relying on a global flag to contain this information. That way, we can remove the whole word and hyphenate it while preserving changes in settings within the word.
Rich Hopkins reported problems with missing sorts when a mortised word is moved to a new line after hitting the justification wall on a previous line. Definitely appears to be related to the setting buffer only having data following the last command sequence.
Using new StyledSort class, the setting buffer keeps the style, letterspacing and hyphenation status, so this problem should go away now. Also, we process the setting buffer less frequently now, such as at justification changes, rather than every time the alphabet is switched (italic to roman, etc).