fix(acceptance): 阻断本地控制面漂移污染验收

本地 K3s 节点此前在 Kubelet 中登记为宿主机资源，负载可挤压 etcd 并在 server 重启后继续污染同一 Run。现在为三节点设置真实可调度预算、独立 etcd/数据卷和锁定依赖镜像，并持续核对 server 与 API/etcd 健康。\n\n每次验收生成独立 Run ID，报告使用独占或原子写入，负载错误记录具体阶段；数据库鉴权不可用返回带 Retry-After 的 503，避免基础设施故障被误报为 401。\n\n验证：Go 全量测试、go vet、gofmt、OpenAPI、bash -n、ShellCheck、报告测试和 k3d 配置解析均通过。

deploy/kubernetes/local-acceptance/k3d.yaml

@@ -4,7 +4,8 @@ metadata:
   name: easyai-acceptance-local
 servers: 3
 agents: 0
-image: rancher/k3s:v1.36.2-k3s1
+image: EASYAI_ACCEPTANCE_K3S_IMAGE
+subnet: EASYAI_ACCEPTANCE_CLUSTER_SUBNET
 kubeAPI:
   host: 127.0.0.1
   hostPort: "16550"
@@ -19,6 +20,21 @@ volumes:
   - volume: EASYAI_ACCEPTANCE_MEDIA_DIR:/var/lib/easyai-acceptance/media
     nodeFilters:
       - server:*
+  - volume: easyai-acceptance-local-server-0-etcd:/var/lib/rancher/k3s/server/db
+    nodeFilters:
+      - server:0
+  - volume: easyai-acceptance-local-server-1-etcd:/var/lib/rancher/k3s/server/db
+    nodeFilters:
+      - server:1
+  - volume: easyai-acceptance-local-server-2-etcd:/var/lib/rancher/k3s/server/db
+    nodeFilters:
+      - server:2
+  - volume: easyai-acceptance-local-server-0-storage:/var/lib/rancher/k3s/storage
+    nodeFilters:
+      - server:0
+  - volume: easyai-acceptance-local-server-1-storage:/var/lib/rancher/k3s/storage
+    nodeFilters:
+      - server:1
 options:
   k3s:
     extraArgs:
@@ -48,3 +64,20 @@ options:
       - arg: --node-taint=easyai.io/control-plane-only=true:NoSchedule
         nodeFilters:
           - server:2
+      - arg: --kubelet-arg=system-reserved=cpu=EASYAI_ACCEPTANCE_SITE_SYSTEM_CPU,memory=EASYAI_ACCEPTANCE_SITE_SYSTEM_MEMORY
+        nodeFilters:
+          - server:0
+          - server:1
+      - arg: --kubelet-arg=kube-reserved=cpu=250m,memory=512Mi
+        nodeFilters:
+          - server:0
+          - server:1
+      - arg: --kubelet-arg=system-reserved=cpu=EASYAI_ACCEPTANCE_WITNESS_SYSTEM_CPU,memory=EASYAI_ACCEPTANCE_WITNESS_SYSTEM_MEMORY
+        nodeFilters:
+          - server:2
+      - arg: --kubelet-arg=kube-reserved=cpu=250m,memory=512Mi
+        nodeFilters:
+          - server:2
+      - arg: --kubelet-arg=eviction-hard=memory.available<512Mi,nodefs.available<10%
+        nodeFilters:
+          - server:*