refactor(ops-cleanup): 拆分 executor + table-driven + 提取常量 + 补测试

代码审查反馈：

1. 文件行数超标：ops_cleanup_service.go 594→413 行。
   拆 opsCleanupPlan / deleteOldRowsByID / truncateOpsTable / isMissingRelationError
   + counts struct 到 ops_cleanup_executor.go (164 行)。

2. runCleanupOnce 89 行→30 行（table-driven）：
   用 []opsCleanupTarget 循环替代三组重复的 opsCleanupPlan → runOne → assign。

3. 魔法值提取常量：
   opsCleanupDefaultSchedule / opsCleanupBatchSize / opsCleanupCronStopTimeout /
   opsCleanupRunTimeout / opsCleanupHeartbeatTimeout。
   ops_settings.go 中 "0 2 * * *" 也统一引用 opsCleanupDefaultSchedule。

4. 补 5 个缺失测试：
   - Reload 未 Start 时 no-op
   - Reload 已 Stop 后 no-op
   - cleanupReloader==nil 时 Update 不 panic
   - Start 重复调用幂等
   - refreshEffectiveBeforeRun 正确更新 snapshot

backend/internal/service/ops_cleanup_overlay_test.go

@@ -194,3 +194,64 @@ func TestUpdateOpsAdvancedSettings_TriggersReload(t *testing.T) {
 		t.Fatalf("expected reloader.Reload called once, got %d", reloader.calls)
 	}
 }
+
+func TestReload_BeforeStart_IsNoop(t *testing.T) {
+	svc := &OpsCleanupService{}
+	if err := svc.Reload(context.Background()); err != nil {
+		t.Fatalf("Reload before Start should return nil, got %v", err)
+	}
+}
+
+func TestReload_AfterStop_IsNoop(t *testing.T) {
+	svc := &OpsCleanupService{started: true, stopped: true}
+	if err := svc.Reload(context.Background()); err != nil {
+		t.Fatalf("Reload after Stop should return nil, got %v", err)
+	}
+}
+
+func TestUpdateOpsAdvancedSettings_NilReloader_NoPanic(t *testing.T) {
+	repo := newRuntimeSettingRepoStub()
+	svc := &OpsService{settingRepo: repo}
+	// cleanupReloader intentionally nil
+
+	cfg := defaultOpsAdvancedSettings()
+	cfg.DataRetention.ErrorLogRetentionDays = 7
+
+	// should not panic
+	if _, err := svc.UpdateOpsAdvancedSettings(context.Background(), cfg); err != nil {
+		t.Fatalf("update with nil reloader: %v", err)
+	}
+}
+
+func TestStart_IdempotentSecondCall(t *testing.T) {
+	svc := &OpsCleanupService{started: true}
+	svc.Start() // second call should be noop, not panic
+}
+
+func TestRefreshEffectiveBeforeRun_UpdatesSnapshot(t *testing.T) {
+	repo := newRuntimeSettingRepoStub()
+	base := config.OpsCleanupConfig{
+		Enabled:               true,
+		Schedule:              "0 2 * * *",
+		ErrorLogRetentionDays: 30,
+	}
+	svc := makeOverlayService(repo, base)
+	svc.computeEffectiveLocked(context.Background())
+
+	if svc.effective.ErrorLogRetentionDays != 30 {
+		t.Fatalf("initial retention should be 30, got %d", svc.effective.ErrorLogRetentionDays)
+	}
+
+	// simulate UI change
+	writeAdvancedSettings(t, repo, OpsDataRetentionSettings{
+		CleanupEnabled:        true,
+		CleanupSchedule:       "0 * * * *",
+		ErrorLogRetentionDays: 7,
+	})
+
+	svc.refreshEffectiveBeforeRun(context.Background())
+	snap := svc.snapshotEffective()
+	if snap.ErrorLogRetentionDays != 7 {
+		t.Fatalf("after refresh, retention should be 7, got %d", snap.ErrorLogRetentionDays)
+	}
+}