Result of splitting a Word at one of its SHY break offsets.
prefix.text already includes a trailing U+002D HYPHEN-MINUS and
its width_pt is the post-shape advance sum (including the
hyphen). suffix.text carries the remaining bytes with
shy_break_offsets rebased to the suffix’s local indexing and
boundary offsets (0 / len) dropped.
Inline item emitted by collect_words. The greedy line-breaker
(and, later, the Knuth-Plass breaker) walks the stream and emits
page geometry; HardBreak is a sentinel that forces a flush of
the in-progress line without contributing any glyphs.
Strip U+00AD (soft hyphen) codepoints from text and return the
stripped string plus the byte offsets in the stripped output
where each SHY originally sat. The offsets mark the codepoint
boundary after the preceding cluster: a break taken at offset
o leaves bytes [0..o) on the previous line and [o..) on the
next.
Try to break word at the latest SHY offset whose prefix-plus-
visible-hyphen fits in max_prefix_width. Returns None if no
valid offset fits. Offsets equal to 0 or word.text.len()
(leading / trailing SHY) are ignored, matching the rule that a
break must produce a non-empty visible prefix and a non-empty
suffix. Consecutive duplicate offsets (e.g. a\u{AD}\u{AD}b →
[1, 1]) are deduped on the fly.