The co-op integration test picks a band the AI has not taken and waits for b.start (#146)

Fixes the flaky CI run reported in #144: a fixed seed, a free band from the public draft with the pick asserted, and the battle start awaited. Closes #144. Contributed by @CELLOPOI.

docs/META.md

@@ -750,6 +750,8 @@ round was over. The official 1 s `broadcastBeginDelay` is not modelled.
 
 ## 6. Testing & tools
 
+* `test/match/lobby-integration.test.js` 的联机轮选使用固定种子 `69`，覆盖 AI 先选走默认策略的情况（GitHub #144）。模拟玩家从当前公开选牌结果中选择未被占用的策略，并断言请求成功；不依赖拒绝后的超时分配。单人和联机用例均通过 `b.start` 确认战斗开始，避免瞬间结束的模拟战斗被公开状态节流而漏掉 `COMBAT` 阶段。
+
 * `test/match/fakeBattle.js` — scriptable DESIGN §5.1 Battle (`FakeBattle.script = (battle) => plan`); inject with
   `new Match({ …, BattleClass: FakeBattle })`.
 * `test/match/harness.js` — `makeMatch(opts)` (virtual scheduler, captured frames, `drive()` for human decisions,