feat: enable document orientation detection for scanned PDFs

- Enable PP-StructureV3's use_doc_orientation_classify feature
- Detect rotation angle from doc_preprocessor_res.angle
- Swap page dimensions (width <-> height) for 90°/270° rotations
- Output PDF now correctly displays landscape-scanned content

Also includes:
- Archive completed openspec proposals
- Add simplify-frontend-ocr-config proposal (pending)
- Code cleanup and frontend simplification

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

backend/app/services/ocr_service.py

@@ -10,6 +10,7 @@ from typing import Any, Dict, List, Optional, Tuple, Union
 from datetime import datetime
 import uuid
 import gc  # For garbage collection
+import warnings  # For suppressing PaddleX deprecation warnings
 
 from paddleocr import PaddleOCR, PPStructureV3
 from PIL import Image
@@ -34,7 +35,21 @@ from app.services.layout_preprocessing_service import (
     get_layout_preprocessing_service,
     LayoutPreprocessingService,
 )
-from app.schemas.task import PreprocessingModeEnum, PreprocessingConfig, TableDetectionConfig
+from app.schemas.task import PreprocessingModeEnum, PreprocessingConfig
+from dataclasses import dataclass
+
+
+@dataclass
+class TableDetectionConfig:
+    """Internal table detection configuration for OCR service.
+
+    Note: This was previously in app.schemas.task but is now internal to OCR service
+    as frontend no longer configures these options.
+    """
+    enable_wired_table: bool = True
+    enable_wireless_table: bool = True
+    enable_region_detection: bool = True
+
 
 # Import dual-track components
 try:
@@ -798,7 +813,12 @@ class OCRService:
                 if textline_ori_model:
                     pp_kwargs['textline_orientation_model_name'] = textline_ori_model
 
-                self.structure_engine = PPStructureV3(**pp_kwargs)
+                # Suppress DeprecationWarning during PPStructureV3 initialization
+                # Workaround for PaddleX bug: it incorrectly treats Python's datetime.utcnow()
+                # deprecation warning as a model loading error in PP-Chart2Table
+                with warnings.catch_warnings():
+                    warnings.filterwarnings('ignore', category=DeprecationWarning)
+                    self.structure_engine = PPStructureV3(**pp_kwargs)
 
                 # Track model loading for cache management
                 self._model_last_used['structure'] = datetime.now()
@@ -881,7 +901,10 @@ class OCRService:
                     if settings.textline_orientation_model_name:
                         cpu_kwargs['textline_orientation_model_name'] = settings.textline_orientation_model_name
 
-                    self.structure_engine = PPStructureV3(**cpu_kwargs)
+                    # Suppress DeprecationWarning during PPStructureV3 initialization (CPU fallback)
+                    with warnings.catch_warnings():
+                        warnings.filterwarnings('ignore', category=DeprecationWarning)
+                        self.structure_engine = PPStructureV3(**cpu_kwargs)
                     self._current_layout_model = layout_model  # Track current model for recreation check
                     # Track table detection config for recreation check
                     if table_detection_config:
@@ -1429,6 +1452,22 @@ class OCRService:
                     raw_ocr_regions=text_regions  # For table content rebuilding
                 )
 
+            # Get detected rotation from layout analysis (default: "0" = no rotation)
+            detected_rotation = "0"
+            if layout_data:
+                detected_rotation = layout_data.get('detected_rotation', '0')
+
+            # Adjust page dimensions based on detected rotation
+            # When rotation is 90° or 270°, the page orientation changes (portrait <-> landscape)
+            # PP-StructureV3 returns coordinates based on the rotated image, so we need to swap dimensions
+            if detected_rotation in ['90', '270']:
+                original_width, original_height = ocr_width, ocr_height
+                ocr_width, ocr_height = original_height, original_width
+                logger.info(
+                    f"Page dimensions adjusted for {detected_rotation}° rotation: "
+                    f"{original_width}x{original_height} -> {ocr_width}x{ocr_height}"
+                )
+
             # Generate Markdown
             markdown_content = self.generate_markdown(text_regions, layout_data)
 
@@ -1450,7 +1489,8 @@ class OCRService:
                 'ocr_dimensions': {
                     'width': ocr_width,
                     'height': ocr_height
-                }
+                },
+                'detected_rotation': detected_rotation  # Document orientation: "0", "90", "180", "270"
             }
 
             # If layout data is enhanced, add enhanced results for converter
@@ -1705,7 +1745,8 @@ class OCRService:
                         'total_elements': result['total_elements'],
                         'reading_order': result['reading_order'],
                         'element_types': result.get('element_types', {}),
-                        'enhanced': True
+                        'enhanced': True,
+                        'detected_rotation': result.get('detected_rotation', '0')  # Document orientation
                     }
 
                     # Extract images metadata