fix(worker): 记住轮询间隔内的容量需求

异步视频任务在 River 轮询空档会释放本地 lease，采样瞬间可能观察不到已经触顶的领取需求，导致自适应容量长期停在初始值。本提交记录采样窗口内出现过的容量饱和，同时保持空闲 Worker 不盲目升容。\n\n移植时保留主线现有的单节点容量阶梯、请求下限和严格失败门禁，仅合入失败路径先记录并排空任务与回调、再删除模拟器资源的清理顺序，并补充脚本守卫。\n\n验证：go test ./... -count=1；go vet ./...；bash -n；ShellCheck；production-acceptance-script-test.sh。

tests/release/production-acceptance-script-test.sh

@@ -247,6 +247,25 @@ for invalid_capacity_state in \
   fi
 done
 
+# shellcheck disable=SC2016 # Match the literal arithmetic expression in the target script.
+grep -Fq 'requests=$((slots * 6))' "$script"
+if grep -Fq '(( requests < 64 ))' "$script"; then
+  echo 'single-node video ladder still overdrives low-capacity steps with a fixed request floor' >&2
+  exit 1
+fi
+
+awk '
+  /^cleanup\(\)/ { capture = 1 }
+  capture { print }
+  capture && /^}/ { exit }
+' "$script" >"$tmp/cleanup-function.sh"
+mark_failed_line=$(grep -n 'mark_run_failed' "$tmp/cleanup-function.sh" | cut -d: -f1)
+cleanup_resources_line=$(grep -n 'cleanup_single_node_kubernetes_resources' "$tmp/cleanup-function.sh" | cut -d: -f1)
+[[ -n $mark_failed_line && -n $cleanup_resources_line && $mark_failed_line -lt $cleanup_resources_line ]] || {
+  echo 'single-node cleanup removes the callback collector before failed outbox work drains' >&2
+  exit 1
+}
+
 passing_pressure_row='2026-08-01T00:00:00Z,0,0,899,149,200,84,1535,0,32,32,0,0,0,0,2,48,0,8,8,0,0,0,16,16,0,0,1535,100'
 pressure_row_passes_live_gates "$passing_pressure_row"
 [[ $(pressure_row_live_gate_reason '2026-08-01T00:00:00Z,0,0,900,149,200,79,1535,0,32,32,0,0,0,0,2,48,0,8,8,0,0,0,16,16,0,0,1535,100') == \