chore: reduce layout threshold to 0.2, tune vLLM memory, and add test scripts and reports

backend/config/vllm_config.yaml

@@ -1,6 +1,6 @@
 # vLLM backend tuning for paddleocr genai_server
 # Docs: https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/PaddleOCR-VL.html#331-server-side-parameter-adjustment
-gpu-memory-utilization: 0.6
+gpu-memory-utilization: 0.35
 max-num-seqs: 4
 enforce-eager: true
 max-model-len: 2048