fix: don't treat bullet-marker columns as real text columns

On PDFs where every list item starts with ● at the left margin and content
at a fixed offset, histogram column detection sees the gap between marker
and content as a gutter and splits each line across two phantom "columns,"
scrambling the reading order (Anthropic's Mythos system card p.73–74).

- layout: reject gutter candidates where the smaller side is ≥80%
  standalone bullet-marker glyphs (•, ●, ○, ◦, ▪, ▫, ◆, ◇, ■, □)
- markdown/classify: add starts_with_bullet_marker helper (narrower than
  is_list_item — excludes numbered/lettered patterns like 1. and a) so
  numbered section headings stay as headings)
- markdown/convert: skip heuristic heading detection on lines that start
  with a bullet marker
- markdown/classify: strip a leading bullet wrapped in a bold/italic run
  (e.g. "**● Label:**" → "- **Label:**") — some PDFs put the marker inside
  the same bold run as the label

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

CLAUDE.md

@@ -61,8 +61,7 @@ src/
 
 - **Unit tests**: inline `#[cfg(test)] mod tests` in each module with synthetic data.
 - **Integration tests**: `tests/integration_tests.rs` with fixture PDFs in `tests/fixtures/`.
-- **Regression suite**: sibling repo `pdf-evals` with 187+ snapshot PDFs. Run `cargo build --release` then `bench.py test` in that repo before committing.
-- **Semantic quality**: run `bench.py score` in `pdf-evals` for the semantic verdict (TEDS + MHS + reading order + char/word + list preservation, composited). Character-level diff alone misclassifies structural improvements (e.g., column-detection rewrites) as regressions — `score` is the tie-breaker. See `pdf-evals/CLAUDE.md` "Semantic scoring".
+- **Regression suite**: sibling repo `pdf-evals` with 179+ snapshot PDFs. Run `cargo build --release` then `bench.py test` in that repo before committing.
 
 ## Debugging
 

src/markdown/convert.rs

@@ -586,27 +586,7 @@ pub(super) fn to_markdown_from_lines_with_tables_and_images(
             .as_ref()
             .and_then(struct_role_heading_level)
             .filter(|level| !overused_heading_levels.contains(level));
-
-        // Protect wrapped list items: when inside a list, a visually-continuing
-        // line (same indent, line-wrap spacing) must not be reclassified as a
-        // heading by the font heuristic — PDFs often bold the lead phrase of a
-        // list item across multiple wrap lines, and an all-bold middle line
-        // would otherwise split one item into a heading + stray body text.
-        // We gate on the document's paragraph threshold so genuine section
-        // headings that follow a numbered paragraph (y_gap > para_threshold)
-        // remain detectable.
-        let looks_like_list_continuation = in_list
-            && match (last_list_x, line.items.first().map(|i| i.x)) {
-                (Some(list_x), Some(curr_x)) => {
-                    let x_ok = curr_x >= list_x - 5.0 && curr_x <= list_x + 50.0;
-                    let y_ok = y_gap >= 0.0 && y_gap <= para_threshold;
-                    x_ok && y_ok && !is_list_item(plain_trimmed)
-                }
-                _ => false,
-            };
-
         let heuristic_heading = if options.detect_headers
-            && !looks_like_list_continuation
             && plain_trimmed.len() > 3
             && plain_trimmed.split_whitespace().count() <= 15
             && !starts_with_bullet_marker(plain_trimmed)
@@ -1473,70 +1453,4 @@ mod tests {
             overused
         );
     }
-
-    #[test]
-    fn test_wrapped_bold_lead_in_list_item_not_heading() {
-        // Regression: numbered-list items whose bold "lead" phrase wraps onto
-        // a second line (e.g. definitions in system cards) must not have the
-        // wrapped line reclassified as a heading. The middle line is
-        // all_bold + standalone (in_paragraph=false while in_list), which
-        // previously tripped the rarity heuristic and emitted #### in the
-        // middle of the item, splitting the body into stray bullets.
-        let make = |text: &str, x: f32, y: f32, bold: bool| {
-            let mut item = make_item(text, 1, None);
-            item.x = x;
-            item.y = y;
-            item.is_bold = bold;
-            item
-        };
-
-        let lines = vec![
-            // "1. **bold lead phrase start**"
-            make_line(vec![
-                make("1. ", 72.0, 700.0, false),
-                make(
-                    "Chemical and biological weapons threat model 1 (CB-1): Non-novel",
-                    90.0,
-                    700.0,
-                    true,
-                ),
-            ]),
-            // wrapped continuation of the bold lead — all_bold, same indent
-            make_line(vec![make(
-                "chemical/biological weapons production capabilities: A model has CB-1",
-                90.0,
-                686.0,
-                true,
-            )]),
-            // body text of the same list item
-            make_line(vec![make(
-                "capabilities if it has the ability to significantly help individuals.",
-                90.0,
-                672.0,
-                false,
-            )]),
-        ];
-
-        let md = to_markdown_from_lines_with_tables_and_images(
-            lines,
-            MarkdownOptions::default(),
-            HashMap::new(),
-            HashMap::new(),
-            &std::collections::HashSet::new(),
-            None,
-        );
-
-        assert!(
-            !md.contains("#### "),
-            "wrapped bold lead must not become a heading: {md}"
-        );
-        assert!(
-            md.lines().filter(|l| l.starts_with("- ")).count() == 0,
-            "continuation body must not become a stray bullet: {md}"
-        );
-        assert!(
-            md.contains("1. ") && md.contains("A model has CB-1"),
-            "numbered list item should remain intact: {md}"
-        );
-    }
 }