fix: 采集失败不降级有数据分行+任务领取计数对齐实际分发行数

- 采集失败时按 latest_likes 分流：无数据才标 failed，已有历史数据保持
  success 并注明沿用上次数据，collection_updated_at 不动；run 行照实记失败
- 种子 seedIfEmpty 末尾统一对账 claimed_quantity，杜绝演示数据口径漂移
- 新增 mysql/0021 对账迁移，修复 0020 物理合并遗留的生产计数漂移
- 运营端展示全面改读实时 dist_total（任务卡双面板/项目树/概览/进度），
  project-service 查询补 dist_total 子查询
- 走查实证：卡片 7/18 vs 详情已领取 2 两处口径打架；保存采集计划补采
  失败会把历史成功行改写为 failed

tests/collection-schedule.test.mjs

@@ -1,4 +1,5 @@
 import assert from "node:assert/strict";
+import { readFile } from "node:fs/promises";
 import test from "node:test";
 import { isCollectionScheduleDue } from "../lib/collection-service.ts";
 import { sortWithNullsLast } from "../lib/sort-utils.ts";
@@ -56,3 +57,24 @@ test("sorts both directions while keeping missing values last", () => {
     ["high", "middle", "low", "missing-a", "missing-b"],
   );
 });
+
+test("a failed collection keeps historically successful rows on success", async () => {
+  const source = await readFile(
+    new URL("../lib/collection-service.ts", import.meta.url),
+    "utf8",
+  );
+
+  // 失败写入必须按 latest_likes 分流：无数据 → failed；
+  // 已有历史数据 → 保持 success，仅描述里说明沿用上次数据。
+  assert.match(
+    source,
+    /collection_status = CASE\s+WHEN latest_likes IS NULL THEN 'failed'\s+ELSE 'success'\s+END/,
+  );
+  assert.match(
+    source,
+    /ELSE \?\s+END,\s+collection_updated_at = CASE\s+WHEN latest_likes IS NULL THEN CURRENT_TIMESTAMP\s+ELSE collection_updated_at/,
+  );
+  assert.match(source, /`\$\{message\}，沿用上次采集数据`/);
+  // run 行照实记录失败，保留可观测性
+  assert.match(source, /UPDATE collection_runs\s+SET status = 'failed'/);
+});