feat: require offline PaddleOCR model assets

backend/tests/test_parsing.py

@@ -6,7 +6,7 @@ from docx import Document
 from pptx import Presentation
 from pptx.util import Inches
 
-from wisegrade.parsing import parse_document
+from wisegrade.parsing import _ocr_engine, parse_document
 
 
 def test_pdf_pages_and_images(tmp_path):
@@ -20,6 +20,11 @@ def test_pdf_pages_and_images(tmp_path):
     assert (tmp_path / "assets" / "page-1.png").exists()
 
 
+def test_scanned_pdf_requires_offline_ocr_package(tmp_path):
+    with pytest.raises(RuntimeError, match="离线OCR模型目录"):
+        _ocr_engine(str(tmp_path / "missing-ocr"))
+
+
 @pytest.mark.skipif(not shutil.which("libreoffice"), reason="LibreOffice is installed in the deployment image")
 def test_docx_and_pptx_render(tmp_path):
     word = Document()