fix: multi-worker translation status and OCR fallback handling

Translation status (multi-worker support):
- Add filesystem lock files (.translating) to track in-progress translations
- Check lock files in /status API when job_state not found in current worker
- Remove lock files on translation success or failure

OCR fallback fix:
- Fix empty pages when layout analysis fails but OCR succeeds
- Change 'enhanced_results' in ocr_results to ocr_results.get('enhanced_results')
- This ensures fallback to text_regions when enhanced_results is empty list

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

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

backend/app/services/ocr_to_unified_converter.py

@@ -439,14 +439,15 @@ class OCRToUnifiedConverter:
         ocr_dimensions = ocr_results.get('ocr_dimensions', {})
 
         # Check if we have enhanced results from PPStructureEnhanced
-        if 'enhanced_results' in ocr_results:
+        # Note: Must check for non-empty list, not just key existence (key may exist with empty list)
+        if ocr_results.get('enhanced_results'):
             pages = self._extract_from_enhanced_results(
                 ocr_results['enhanced_results'],
                 raw_text_regions=raw_text_regions,
                 ocr_dimensions=ocr_dimensions
             )
         # Check for traditional OCR results with text_regions at top level (from process_file_traditional)
-        elif 'text_regions' in ocr_results:
+        elif ocr_results.get('text_regions'):
             pages = self._extract_from_traditional_ocr(ocr_results)
         # Check for traditional layout_data structure
         elif 'layout_data' in ocr_results: