fix: 推理/评测结果同步、数据集统计与算力节点管理增强

后端:
- 抽取 fetch_eval_result_content 复用函数，model_eval_detail 直接应用评测任务结果
- health 接口移除数据库依赖，返回静态指标
- 数据集: count_dataset_records JSON 感知计数; 文件统计改为从 dataset_files 聚合重算; 在线编辑记录 size/record_count/version_no; 上传同步批处理
- 算力节点: 调度支持 requested GPU 子集校验与容量计算; 新增 delete_compute_node(含活动任务保护)及 DELETE 接口; 连接池 connect_timeout
- 评测任务落库 basic_metrics/score/completed_time, failed/stopped 记录 error

评测引擎:
- _load_dataset 支持 JSON/JSONL 文件
- 新增 exact match 与文本相似度指标, 余弦相似度去掉 2 样本限制

前端:
- 算力节点列表「维护」改为「删除」(带确认弹窗), compute.ts 新增 deleteComputeNode
- 数据集上传超时调整为 120s; FineTuneTask 增加 compute_node_id; GpuInfo 状态增加 reserved

frontend/src/views/eval/EvalView.vue

@@ -24,14 +24,16 @@ const leaderboard = ref([
   { rank: 3, name: 'Qwen-Max', score: 85.3 },
 ])
 
-async function loadEvalList() {
-  evalLoading.value = true
+const ACTIVE_STATUSES = new Set(['pending', 'queued', 'running'])
+
+async function loadEvalList(options: { silent?: boolean } = {}) {
+  if (!options.silent) evalLoading.value = true
   try {
     evalList.value = (await getEvalList()) || []
   } catch {
     evalList.value = []
   } finally {
-    evalLoading.value = false
+    if (!options.silent) evalLoading.value = false
   }
 }
 
@@ -56,14 +58,21 @@ function handleViewDetail(row: any) {
 }
 
 const { start: startPolling, stop: stopPolling } = usePolling(
-  () => loadEvalList(),
+  async () => {
+    await loadEvalList({ silent: true })
+    if (!evalList.value.some((item) => ACTIVE_STATUSES.has(String(item.status || '')))) {
+      stopPolling()
+    }
+  },
   5000,
   { immediate: false },
 )
 
 onMounted(async () => {
   await loadEvalList()
-  startPolling()
+  if (evalList.value.some((item) => ACTIVE_STATUSES.has(String(item.status || '')))) {
+    startPolling()
+  }
 })
 
 onUnmounted(() => {