Improve the handling of punctuation in Undetermined

[joanise]

Right now, the Undetermined und mapping issues errors if it tries to process a word with a colon or other punctuation in it. The comma and apostrophe have a sound mapping, but not the colon, semi-colon or period. Punctuation should normally be tokenized out of the words to g2p, but sometimes they are not, e.g., when we're using an und fallback for text in a language that has that punctuation symbol as a latter.

When a punctuation symbol is passed to und for g2p mapping, we should do something reasonable. Colon could logically map to the length marker, and other punctuation marks could logically just be stripped out.