feat: add pass 3 text-line regression to deskew pipeline

After iterative projection (pass 1) and word-alignment (pass 2), a third
pass uses Tesseract word positions + linear regression per text line to
measure and correct residual rotation. This catches cases where passes 1-2
leave significant slope (e.g. 1.7° residual on heavily skewed scans).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

klausur-service/backend/vocab_worksheet_api.py

@@ -1371,14 +1371,16 @@ async def _run_ocr_pipeline_for_page(
     except Exception as e:
         logger.warning(f"Could not create pipeline session in DB: {e}")
 
-    # 3. Two-pass deskew: iterative (±5°) + word-alignment residual
+    # 3. Three-pass deskew: iterative + word-alignment + text-line regression
     t0 = _time.time()
     deskewed_bgr, angle_applied, deskew_debug = deskew_two_pass(img_bgr.copy())
     angle_pass1 = deskew_debug.get("pass1_angle", 0.0)
     angle_pass2 = deskew_debug.get("pass2_angle", 0.0)
+    angle_pass3 = deskew_debug.get("pass3_angle", 0.0)
 
-    logger.info(f"  deskew: pass1={angle_pass1:.2f} pass2={angle_pass2:.2f} "
-                f"total={angle_applied:.2f} ({_time.time() - t0:.1f}s)")
+    logger.info(f"  deskew: p1={angle_pass1:.2f} p2={angle_pass2:.2f} "
+                f"p3={angle_pass3:.2f} total={angle_applied:.2f} "
+                f"({_time.time() - t0:.1f}s)")
 
     # 4. Dewarp
     t0 = _time.time()