feat: parse document result packages from middle json only

Content comes from the middle json and its paired markdown; the content list is deliberately ignored because it escapes markdown characters. The external-data exception boundary covers both reading and structure parsing so no native exception escapes the OCR client. The layout parser is unchanged.

tests/unit/test_monkey_ocr.py

@@ -190,7 +190,9 @@ class TestSniffMedia:
             pytest.param(b"GIF87a" + b"0" * 20, "image", id="gif87a"),
             pytest.param(b"RIFF\x00\x00\x00\x00WEBPVP8 ", "image", id="webp"),
             # 图片魔数优先: 含 XMP 元数据的图片可能在 1KB 内带 `%PDF-` 字样
-            pytest.param(b"\xff\xd8\xff\xe0" + b"<xmp>%PDF-</xmp>", "image", id="jpeg-mentions-pdf"),
+            pytest.param(
+                b"\xff\xd8\xff\xe0" + b"<xmp>%PDF-</xmp>", "image", id="jpeg-mentions-pdf"
+            ),
             pytest.param(b"<html><body>paywall</body></html>", "unknown", id="html"),
             pytest.param(b"GARBAGE-NOT-AN-IMAGE", "unknown", id="garbage"),
             pytest.param(b"", "unknown", id="empty"),
@@ -205,7 +207,9 @@ class TestSniffMedia:
         assert _upload_part(_PDF_BODY) == ("doc.pdf", _PDF_BODY, "application/pdf")
         assert _upload_part(_JUNK + _PDF_BODY)[0] == "doc.pdf"
 
-    @pytest.mark.parametrize("data", [b"\xff\xd8\xff\xe0abc", b"\x89PNG\r\n\x1a\nabc", b"junk", b""])
+    @pytest.mark.parametrize(
+        "data", [b"\xff\xd8\xff\xe0abc", b"\x89PNG\r\n\x1a\nabc", b"junk", b""]
+    )
     def test_everything_else_keeps_the_legacy_label(self, data):
         """非 PDF 路径字节级不变: 服务对图片按内容解码(2026-10-21 实测 PNG/TIFF/BMP 标 image.jpg 成功)。"""
         assert _upload_part(data) == ("image.jpg", data, "image/jpeg")