refactor: restructure services and add image/pdf extraction utilities

- Add image_extractor, pdf_downloader, pi_client, trends services
- Add shared utils module
- Refactor summarizer, embedder, routes for cleaner separation
- Update tests to match new service structure

tests/test_admin_phase4.py

@@ -141,7 +141,7 @@ class TestCleanupTmp:
         old_mtime = time.time() - 25 * 3600
         os.utime(old_dir, (old_mtime, old_mtime))
 
-        monkeypatch.setattr("app.services.cleaner._TMP_DIR", tmp_dir)
+        monkeypatch.setattr("app.services.cleaner.TMP_DIR", tmp_dir)
         from app.services.cleaner import cleanup_tmp
         result = cleanup_tmp()
 
@@ -158,7 +158,7 @@ class TestCleanupTmp:
         recent_dir.mkdir()
         (recent_dir / "paper.pdf").write_text("fake pdf")
 
-        monkeypatch.setattr("app.services.cleaner._TMP_DIR", tmp_dir)
+        monkeypatch.setattr("app.services.cleaner.TMP_DIR", tmp_dir)
         from app.services.cleaner import cleanup_tmp
         result = cleanup_tmp()
 
@@ -168,7 +168,7 @@ class TestCleanupTmp:
 
     def test_cleanup_empty_dir(self, tmp_path, monkeypatch):
         """data/tmp/ 不存在时安全返回。"""
-        monkeypatch.setattr("app.services.cleaner._TMP_DIR", tmp_path / "nonexistent")
+        monkeypatch.setattr("app.services.cleaner.TMP_DIR", tmp_path / "nonexistent")
         from app.services.cleaner import cleanup_tmp
         result = cleanup_tmp()
         assert result["scanned"] == 0
@@ -187,7 +187,7 @@ class TestCleanupTmp:
         recent_dir = tmp_dir / "2401.new"
         recent_dir.mkdir()
 
-        monkeypatch.setattr("app.services.cleaner._TMP_DIR", tmp_dir)
+        monkeypatch.setattr("app.services.cleaner.TMP_DIR", tmp_dir)
         from app.services.cleaner import cleanup_tmp
         result = cleanup_tmp()
 
@@ -318,7 +318,7 @@ class TestDeletePapersByDateRange:
         (papers_dir / "2401.10001").mkdir()
         (papers_dir / "2401.10001" / "meta.json").write_text("{}")
 
-        monkeypatch.setattr("app.services.cleaner._PAPERS_DIR", papers_dir)
+        monkeypatch.setattr("app.services.cleaner.PAPERS_DIR", papers_dir)
 
         result = await delete_papers_by_date_range(
             db_session,