feat: 完成 DOCX R4 元素级视觉门禁

建立 Chromium、Word 与 WPS 的可编辑内容、段落几何、页眉页脚、封面、媒体和逐页视觉比较链路。

支持封面独立分节、正文页码重启、主题页边距与横向纸张，并将自然分页差异降为诊断项。

修正代码块内边距和折叠表格单元格边框映射，14 套主题生产矩阵与六组布局视觉矩阵通过。

packages/document-visual-diff/src/text.ts

@@ -1,10 +1,17 @@
 import type {
+  PdfDocumentSnapshot,
+  PdfEditableContentLine,
   PdfPointBounds,
   PdfTextItemSnapshot,
   PdfTextLineSnapshot,
+  VisualPageSemanticExpectation,
 } from "./types.js";
 
 const ZERO_WIDTH_AND_CONTROL = /[\u0000-\u0008\u000b\u000c\u000e-\u001f\u007f\u200b-\u200d\u2060\ufeff]/gu;
+const CJK_RADICAL_VARIANTS = new Map([
+  ["⺠", "民"],
+  ["⻔", "门"],
+]);
 const PAGE_NUMBER_PATTERNS = [
   /^(?:[-—–]\s*)?\d+(?:\s*[/／]\s*\d+)?(?:\s*[-—–])?$/u,
   /^第\s*\d+\s*页(?:\s*(?:[/／]|共)\s*\d+\s*页)?$/u,
@@ -14,6 +21,10 @@ const PAGE_NUMBER_PATTERNS = [
 export function normalizePdfText(value: string): string {
   return value
     .normalize("NFKC")
+    .replace(/戶/gu, "户")
+    .replace(/[\u2e80-\u2eff]/gu, (character) =>
+      CJK_RADICAL_VARIANTS.get(character) ?? character
+    )
     .replace(ZERO_WIDTH_AND_CONTROL, "")
     .replace(/\u00a0/gu, " ")
     .replace(/[ \t]+/gu, " ")
@@ -24,6 +35,69 @@ export function normalizePdfContentText(value: string): string {
   return normalizePdfText(value).replace(/\s+/gu, "");
 }
 
+export function normalizePdfEditableText(value: string): string {
+  return normalizePdfContentText(value).replace(/[☐☑☒•◦▪\uf0b7]/gu, "");
+}
+
+function isExpectedHeaderLine(
+  line: PdfTextLineSnapshot,
+  pageHeightPt: number,
+  expectation: VisualPageSemanticExpectation | undefined,
+): boolean {
+  if (
+    !expectation?.headerVisible ||
+    line.bounds.y + line.bounds.height / 2 > pageHeightPt * 0.12
+  ) {
+    return false;
+  }
+  const lineText = normalizePdfEditableText(line.normalizedText);
+  return (expectation.headerSlots ?? []).some((slot) => {
+    const slotText = normalizePdfEditableText(slot.text);
+    return slotText.length > 0 && lineText.includes(slotText);
+  });
+}
+
+function isInternalLayoutSpacerLine(line: PdfTextLineSnapshot): boolean {
+  return (
+    normalizePdfEditableText(line.normalizedText) === "." &&
+    line.bounds.height <= 1.5 &&
+    line.items.length === 1
+  );
+}
+
+export function buildPdfEditableContentText(
+  snapshot: PdfDocumentSnapshot,
+  pageSemantics: readonly VisualPageSemanticExpectation[] = [],
+): string {
+  return buildPdfEditableContentLines(snapshot, pageSemantics)
+    .map((item) => item.normalizedText)
+    .join("");
+}
+
+export function buildPdfEditableContentLines(
+  snapshot: PdfDocumentSnapshot,
+  pageSemantics: readonly VisualPageSemanticExpectation[] = [],
+): PdfEditableContentLine[] {
+  return snapshot.pages.flatMap((page) => {
+    const expectation = pageSemantics.find(
+      (item) => item.physicalPageNumber === page.pageNumber,
+    );
+    return page.lines.flatMap((line) => {
+      if (
+        line.role !== "content" ||
+        isExpectedHeaderLine(line, page.heightPt, expectation) ||
+        isInternalLayoutSpacerLine(line)
+      ) {
+        return [];
+      }
+      const normalizedText = normalizePdfEditableText(line.normalizedText);
+      return normalizedText
+        ? [{ pageNumber: page.pageNumber, line, normalizedText }]
+        : [];
+    });
+  });
+}
+
 function unionBounds(items: readonly PdfTextItemSnapshot[]): PdfPointBounds {
   const left = Math.min(...items.map((item) => item.bounds.x));
   const top = Math.min(...items.map((item) => item.bounds.y));