Add Vision-LLM OCR Fusion (Step 4) for degraded scans

New module vision_ocr_fusion.py: Sends scan image + OCR word
coordinates + document type to Qwen2.5-VL 32B. The LLM reads
the image visually while using OCR positions as structural hints.

Key features:
- Document-type-aware prompts (Vokabelseite, Woerterbuch, etc.)
- OCR words grouped into lines with x/y coordinates in prompt
- Low-confidence words marked with (?) for LLM attention
- Continuation row merging instructions in prompt
- JSON response parsing with markdown code block handling
- Fallback to original OCR on any error

Frontend (admin-lehrer Grid Review):
- "Vision-LLM" checkbox toggle
- "Typ" dropdown (Vokabelseite, Woerterbuch, etc.)
- Steps 1-3 defaults set to inactive

Activate: Check "Vision-LLM", select document type, click "OCR neu + Grid".

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

admin-lehrer/components/ocr-pipeline/StepGridReview.tsx

@@ -67,6 +67,10 @@ export function StepGridReview({ sessionId, onNext, saveRef }: StepGridReviewPro
     setOcrMaxCols,
     ocrMinConf,
     setOcrMinConf,
+    visionFusion,
+    setVisionFusion,
+    documentCategory,
+    setDocumentCategory,
     rerunOcr,
   } = useGridEditor(sessionId)
 
@@ -291,6 +295,22 @@ export function StepGridReview({ sessionId, onNext, saveRef }: StepGridReviewPro
           </select>
         </label>
 
+        <span className="text-gray-400 dark:text-gray-500">|</span>
+        <label className="flex items-center gap-1 cursor-pointer" title="Step 4: Vision-LLM Fusion — Qwen2.5-VL korrigiert OCR anhand des Bildes">
+          <input type="checkbox" checked={visionFusion} onChange={(e) => setVisionFusion(e.target.checked)} className="rounded w-3 h-3 accent-orange-500" />
+          <span className={`${visionFusion ? 'text-orange-500 dark:text-orange-400 font-medium' : 'text-gray-500 dark:text-gray-400'}`}>Vision-LLM</span>
+        </label>
+        <label className="flex items-center gap-1" title="Dokumenttyp fuer Vision-LLM Prompt">
+          <span className="text-gray-500 dark:text-gray-400">Typ:</span>
+          <select value={documentCategory} onChange={(e) => setDocumentCategory(e.target.value)} className="px-1 py-0.5 text-xs rounded border border-gray-200 dark:border-gray-600 bg-white dark:bg-gray-700 text-gray-700 dark:text-gray-300">
+            <option value="vokabelseite">Vokabelseite</option>
+            <option value="woerterbuch">Woerterbuch</option>
+            <option value="arbeitsblatt">Arbeitsblatt</option>
+            <option value="buchseite">Buchseite</option>
+            <option value="sonstiges">Sonstiges</option>
+          </select>
+        </label>
+
         <div className="ml-auto flex items-center gap-2">
           <button
             onClick={() => {