sitetile: dialogue — width-aware name cap, whole-run Q&A demotion, quoteRunReport

[cverorg]

What happened

An interview post written one quote block per question (**SODAART：question** over **摩卡麵包：** answer) rendered four of its fourteen questions as chat bubbles — question as the speaker label, answer as the bubble — and the other ten as quotations, on the same page. The 24-character speaker cap counts UTF-16 units and was calibrated on Latin names; in Chinese or Japanese, 24 units is a sentence.

Change

• Width-aware cap: an East Asian wide character is two columns, so 24 columns hold a name in either script and a sentence in neither. RE_TURN_HEAD keeps its literal unit cap (reef-mcp vendors it as a plain regex); the width check is a second gate in dialogueTurn.
• Whole-run judgment: a bold-headed block that is NOT a turn (over the cap, a list beneath it, nothing beneath it) beside blocks that are makes the run a list of same-shaped items (a Q&A, a series nav) — no block in it is a turn. A turn beside an ordinary quotation still mixes, as before.
• quoteRunReport(body) (new export): per quote run, what the renderer would make of it — blocks, bold heads, turns, names, demoted — from the same classifier bodyHtml uses, so the editing door can warn "these would be bubbles, these would not" without a second copy of the rule.
• collectQuoteRun is the run collector factored out of bodyHtml, unchanged.

Tests

Three added (width cap at/over the limit; Q&A demotion with the interview in miniature, plus the turn-beside-quotation shape that must keep mixing; report agreement with rendered output). scripts/test.sh all green, sitetile 193 pass.

Follow-up in CVERInc/reef: re-vendor, and a dialogue-shape warning on save/create/verify built on quoteRunReport.

https://claude.ai/code/session_014rHD92DiWFtyWLRApphqub