fix(worker): 限制大图归一化峰值内存

6K 多参考图在 P24 下会并发进入高内存缩放并击穿 2 GiB Worker。新增独立可配置的图片归一化并发，生产默认 2，只约束解码、缩放和重编码阶段，不占用视频上游等待槽；缩放器改为内存稳定的近似双线性实现。\n\n验证：6K 到 6000x2400 转换、信号量串行化、Go 全量测试、gofmt、Kubernetes 客户端 dry-run。

deploy/kubernetes/production/application.yaml

@@ -66,6 +66,8 @@ spec:
               value: "16"
             - name: AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY
               value: "8"
+            - name: AI_GATEWAY_MEDIA_IMAGE_NORMALIZATION_CONCURRENCY
+              value: "2"
             - name: AI_GATEWAY_MEDIA_OSS_ACCESS_KEY_ID
               valueFrom:
                 secretKeyRef:
@@ -335,6 +337,8 @@ spec:
               value: "16"
             - name: AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY
               value: "8"
+            - name: AI_GATEWAY_MEDIA_IMAGE_NORMALIZATION_CONCURRENCY
+              value: "2"
             - name: AI_GATEWAY_MEDIA_OSS_ACCESS_KEY_ID
               valueFrom:
                 secretKeyRef:
@@ -492,6 +496,8 @@ spec:
               value: "24"
             - name: AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY
               value: "24"
+            - name: AI_GATEWAY_MEDIA_IMAGE_NORMALIZATION_CONCURRENCY
+              value: "2"
             - name: AI_GATEWAY_MEDIA_OSS_ACCESS_KEY_ID
               valueFrom:
                 secretKeyRef:
@@ -649,6 +655,8 @@ spec:
               value: "24"
             - name: AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY
               value: "24"
+            - name: AI_GATEWAY_MEDIA_IMAGE_NORMALIZATION_CONCURRENCY
+              value: "2"
             - name: AI_GATEWAY_MEDIA_OSS_ACCESS_KEY_ID
               valueFrom:
                 secretKeyRef: