Switch Vision-LLM Fusion to llama3.2-vision:11b

qwen2.5vl:32b needs ~100GB RAM and crashes Ollama.
llama3.2-vision:11b is already installed and fits in memory.

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

klausur-service/backend/vision_ocr_fusion.py

@@ -22,7 +22,7 @@ import numpy as np
 logger = logging.getLogger(__name__)
 
 OLLAMA_BASE_URL = os.getenv("OLLAMA_BASE_URL", "http://host.docker.internal:11434")
-OLLAMA_HTR_MODEL = os.getenv("OLLAMA_HTR_MODEL", "qwen2.5vl:32b")
+VISION_FUSION_MODEL = os.getenv("VISION_FUSION_MODEL", "llama3.2-vision:11b")
 
 # Document category → prompt context
 CATEGORY_PROMPTS: Dict[str, Dict[str, str]] = {
@@ -225,7 +225,7 @@ async def vision_fuse_ocr(
             resp = await client.post(
                 f"{OLLAMA_BASE_URL}/api/generate",
                 json={
-                    "model": OLLAMA_HTR_MODEL,
+                    "model": VISION_FUSION_MODEL,
                     "prompt": prompt,
                     "images": [img_b64],
                     "stream": False,