Coalesce <br><br> hard-break runs into a paragraph break

[konard]

Fixes #121.

Summary

JS --capture api (export-html → markdown) emitted trailing two-space CommonMark hard breaks ( \n) on lines that should end paragraphs. Two adjacent hard breaks in the markdown source render as <br><br> inside a single <p> (GitHub, MkDocs, Pandoc), cramming captions against images with no vertical spacing — and the "blank" separator line actually carries trailing whitespace, polluting diffs.

Root cause

Google Docs export-html wraps every visual line break in <br>, including <br><br> at paragraph boundaries:

<p>
  Caption A:<br>
  <img …><br><br>
  Caption B:<br>
  <img …>
</p>

Turndown faithfully emits one trailing-two-space-newline pair for each <br>, producing \n \n between the two captions. CommonMark says two or more adjacent hard breaks still belong to one paragraph, so renderers cram both captions into a single <p>. The Rust html2md crate already coalesces consecutive <br>s correctly, so this is a JS-only bug.

Fix

After Turndown runs, collapse runs of two or more adjacent hard breaks to \n\n:

function coalesceBrRunsToParagraphBreak(markdown) {
  return markdown.replace(/(?: {2,}\n){2,}/g, '\n\n');
}

Applied in both convertHtmlToMarkdown (used by --capture api) and convertHtmlToMarkdownEnhanced. The regex tolerates any post-processing that might leave 2+ trailing spaces on a hard break.

Tests

New regression tests with a shared fixture (paragraph-vs-line-break.html) modelled on the issue's spec — two captions separated by <br><br> inside a single <p>:

• js/tests/unit/paragraph-vs-line-break.test.js — verifies both convertHtmlToMarkdown and convertHtmlToMarkdownEnhanced produce a truly empty separator line (no trailing whitespace) and that markdown-it CommonMark renders the two captions as separate <p>s.
• rust/tests/integration/paragraph_vs_line_break.rs — same checks as a parity guard (Rust converter never had the bug).

Test plan

• cd js && npm test -- tests/unit — 257/261 pass; the 4 failures are pre-existing browser.test.js failures from a missing Playwright Chromium binary in the sandbox, unrelated to this change.
• cd js && npm test -- tests/unit/paragraph-vs-line-break.test.js tests/unit/html2md.test.js tests/unit/postprocess.test.js tests/unit/gdocs.test.js tests/unit/html2md-br-in-list-item.test.js — all pass.
• cd rust && cargo test — 92 unit + 96 integration tests pass, 0 failed.
• cd js && npm run lint — 0 errors.
• cd js && npx prettier --check on changed files — all formatted.
• cd rust && cargo clippy --all-targets -- -D warnings — clean.

Files changed

• js/src/lib.js — coalesceBrRunsToParagraphBreak helper + two call sites.
• js/.changeset/issue-121-paragraph-vs-line-break.md — patch changeset.
• js/tests/fixtures/paragraph-vs-line-break.html, js/tests/unit/paragraph-vs-line-break.test.js — JS regression test (uses markdown-it to verify CommonMark rendering).
• js/package.json, js/package-lock.json, js/yarn.lock — add markdown-it as a dev dependency.
• rust/tests/fixtures/paragraph-vs-line-break.html, rust/tests/integration/paragraph_vs_line_break.rs, rust/tests/integration/mod.rs — Rust parity test.
• rust/Cargo.lock — sync to 0.3.14 (main bumped Cargo.toml without lockfile update).
• experiments/repro-br-paragraph.mjs — reproduction script that pinpointed the bug.

[konard]

Working session summary

Both empty so far. The original until-loop background job is running and will notify on completion. Let me wait passively.

---

This summary was automatically extracted from the AI working session output.

[konard]

🤖 Solution Draft Log

This log file contains the complete execution trace of the AI solution draft process.

💰 Cost: $8.091367

📊 Context and tokens usage:

Claude Opus 4.7: (2 sub-sessions)

1. 116.8K / 1M (12%) input tokens, 33.4K / 128K (26%) output tokens
2. 59.0K / 1M (6%) input tokens, 8.9K / 128K (7%) output tokens

Total: (2.5K new + 157.2K cache writes + 11.8M cache reads) input tokens, 48.7K output tokens, $8.091367 cost

🤖 Models used:

• Tool: Anthropic Claude Code
• Requested: opus
• Model: Claude Opus 4.7 (claude-opus-4-7)

📎 Log file uploaded as Gist (3475KB)

• View complete solution draft log

---

Now working session is ended, feel free to review and add any feedback on the solution draft.

[konard]

✅ Ready to merge

This pull request is now ready to be merged:

• All CI checks have passed
• No merge conflicts
• No pending changes

---

Monitored by hive-mind with --auto-restart-until-mergeable flag