Hyphenation: consider words that contain a NBSP

[ondras]

The main goal here is to fix https://github.com/Kozea/WeasyPrint/issues/2270.

My change is simple:

• adjust the get_next_word_boundaries function so it considers the whole word-nbsp-word cluster as a single potentially-hyphenable word,
• ignore suggested hyphenation dictionary item(s) that end with a nbsp

Tests seem to pass. However: