test(project-lifecycle): align settlement readback fixture

[Duang777]

Summary

• add the required settlement identity and progress fields to the project lifecycle test readback
• keep the fixture on the settled path so the test remains scoped to post-writeback hook replay

Validation

• uvx --from pytest pytest -q tests/cli_commands/test_project_lifecycle_goal_channel.py
• uvx --from ruff==0.15.22 ruff check tests/cli_commands/test_project_lifecycle_goal_channel.py
• uv run loopx canary premerge --from-git-diff
• git diff --check

[Duang777]

Exact-head CI triage for ef440fdd70284184583d94840199fe434aff487c: the completed failures are outside this fixture-only diff.

• node-minimum-compatibility runs two SQLite task_lease_inspection rows on Node 22.18.0 and rejects SQLite 3.50.2. The same failure is present on independent PR fix(ci): restore full public smoke contracts #4860; test(ci): qualify lease inspection on public Node #4857 contains the isolated public-minimum provider qualification and its Node minimum lane passes.
• stage2c (mutants 0) artifact executed 54 mutants and killed 53. The sole survivor was fence_unshared_state_lock: its cross-process Python oracle can pass when Node startup exceeds the fixed 200 ms window. test(ci): stabilize state-lock mutation oracle #4863 binds that unchanged mutant to the existing in-process state-lock oracle; the full suite then kills 54/54 locally, with six focused passes.

No #4859 code change is indicated by either failure.

[huangruiteng]

评审 exact head：ef440fdd70284184583d94840199fe434aff487c

动机

这项修改修复的是一个真实且已经阻塞其他 PR 的测试夹具漂移。生产 QuotaSettlementReadback 已包含 identity 与 progress，handle_project_lifecycle_command 在绑定 Todo/turn 的 refresh-state 路径会调用 attach_settlement_progress；旧夹具却只提供 delivery.failure，导致测试在进入它真正要验证的 post-writeback hook 重放之前就因缺少 progress 崩溃。

我在当前 main 上复现了这一点：同一测试文件为 1 failed / 8 passed，失败点正是 readback.progress。在本 PR head 上则为 9/9 通过。这里最小且正确的修复就是把既有 mock 补到当前生产读回合同，而不是给生产代码加一个容忍畸形测试对象的 fallback。

改动思路

PR 没有改变 settlement、hook 或 CLI 的任何生产决策，只把目标测试中的 SimpleNamespace 补成已结算读回：identity.value=None，以及 quota_settlement_progress_v0 / settled / next_step=None / heartbeat。这样第一次调用仍生成 sidecar intent，第二次调用仍从 durable sidecar 重放，原有 calls == 1、intent_count == 1、replayed_hooks == ['fixture.stage'] 和零外部写断言都能真正执行。

这也保持了正确的所有权：TypeScript settlement readback 与 Python attach_settlement_progress 仍是合同 owner；测试夹具只负责忠实提供该合同，不复制新的结算规则。

具体改动

关键代码讲解

• tests/cli_commands/test_project_lifecycle_goal_channel.py 的 test_refresh_state_dispatches_and_replays_post_writeback_sidecar：给 monkeypatch 的 read_heartbeat_settlement 返回值增加 identity 和 typed progress。这是唯一改动，+8/-1，没有生产代码、持久化格式或公开接口变化。
• progress.state='settled' 很关键：它让测试继续走既有 post-writeback replay 语义，而不是通过构造 pending/failed 状态绕开被测路径。

对主干的风险

风险很低且局限在测试准确性。最强反例是“仅补属性让测试不崩，但没有真正验证重放”；现有后置断言仍覆盖 producer 只调用一次、第二次识别 replay、没有外部写，因此该反例被挡住。

验证结果：

• 当前 main 对照：目标文件 1 failed / 8 passed，稳定复现缺少 progress；
• PR head：目标文件 9 passed；
• Ruff、git diff --check、Python compile 与 risk-based premerge gate 均通过；
• premerge 无 manual hold；当前 main merge-tree 无冲突。

该 PR 不声称证明真实持久化后端；那属于既有 settlement integration 覆盖。就这个 fixture-only 边界而言，没有未解决 blocker。

我的整体评价

这是完整、可回滚、必要的最小修复：它恢复已有 durable regression test 的可达性，同时不把测试兼容逻辑泄漏进生产路径。未来向前看也不需要抽取新的 fixture 框架；单一场景的 typed settled literal 比额外抽象更易读。

结论：APPROVE。此批准只绑定上述 exact head，不授予合并权限。

English verdict: APPROVE - head ef440fd faithfully updates one stale settlement readback fixture; current main reproduces the missing-progress failure, while the reviewed head passes 9 tests, Ruff, diff hygiene, and the risk-based premerge gate.