feat: OCR pipeline step 8 — validation view with image detection & generation

Replaces the stub StepGroundTruth with a full side-by-side Original vs
Reconstruction view. Adds VLM-based image region detection (qwen2.5vl),
mflux image generation proxy, sync scroll/zoom, manual region drawing,
and score/notes persistence.

New backend endpoints: detect-images, generate-image, validate, get validation.
New standalone mflux-service (scripts/mflux-service.py) for Metal GPU generation.
Dockerfile.base: adds fonts-liberation (Apache-2.0).

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

admin-lehrer/app/(admin)/ai/ocr-pipeline/page.tsx

@@ -313,7 +313,7 @@ export default function OcrPipelinePage() {
       case 6:
         return <StepReconstruction sessionId={sessionId} onNext={handleNext} />
       case 7:
-        return <StepGroundTruth />
+        return <StepGroundTruth sessionId={sessionId} onNext={handleNext} />
       default:
         return null
     }

admin-lehrer/app/(admin)/ai/ocr-pipeline/types.ts

@@ -264,6 +264,24 @@ export interface WordGroundTruth {
   notes?: string
 }
 
+export interface ImageRegion {
+  bbox_pct: { x: number; y: number; w: number; h: number }
+  prompt: string
+  description: string
+  image_b64: string | null
+  style: 'educational' | 'cartoon' | 'sketch' | 'clipart' | 'realistic'
+}
+
+export type ImageStyle = ImageRegion['style']
+
+export const IMAGE_STYLES: { value: ImageStyle; label: string }[] = [
+  { value: 'educational', label: 'Lehrbuch' },
+  { value: 'cartoon', label: 'Cartoon' },
+  { value: 'sketch', label: 'Skizze' },
+  { value: 'clipart', label: 'Clipart' },
+  { value: 'realistic', label: 'Realistisch' },
+]
+
 export const PIPELINE_STEPS: PipelineStep[] = [
   { id: 'deskew', name: 'Begradigung', icon: '📐', status: 'pending' },
   { id: 'dewarp', name: 'Entzerrung', icon: '🔧', status: 'pending' },