test(chat): align cadence and exact Todo fixtures

[loopx-agent]

Two Chat regression fixtures still encoded older contracts: the absent cadence catalog expected completed Todos, and the manager exact-Todo mock returned a list. On the current main baseline both tests fail; this change uses the already documented five-effective-Turn default and returns the canonical single todo shape for exact reads.

The existing continuation, conditions, decision-scope and truncation assertions remain. The exact-read fixture now verifies the selected Goal/Todo and retains scope assertions; two negative cases distinguish a missing record from an unavailable authority without exposing its error. Production code, configuration and permissions are unchanged. The separate SSH evidence fixture repair is already proposed in #5961 and is reused rather than duplicated here.

Validation: both failing nodes reproduced on immutable base d33cc133e5948a30d3e5535e0339fd1ed34c2497; 26 tests passed across both changed modules and the real exact-Todo CLI suite. A separate isolated File/SQLite manager read exercised full owner continuation, conditions/scopes, non-owner privacy, missing and invalid selectors, with source bytes unchanged. Ruff, compile and diff checks passed; the native canary ran four direct checks with no selected risk checks or manual holds. These results do not claim the entire repository test suite is healthy.

Acceptance: existing docs/quota-allocation.md cadence default and loopx/control_plane/todos/list_readback.py exact-read contract; no new decision owner or vocabulary. Future-facing pass: keep the selector-aware double local to its consumer test; no new shared fixture abstraction is needed. The negative cases guard existing public read semantics.

Author: model_agent, OpenAI GPT-6.1 Sol. Independent review is still required; no merge is included.

[loopx-agent]

Reviewer: model_agent; gpt-6.1-sol; OpenAI; runtime_reported; reasoning_effort=xhigh.

Approval conclusion (author-owned PR; GitHub blocks formal self-approval)
Exact head: d971dc6; base: d33cc13.

动机

维护者运行 Chat 回归测试时，会遇到已落后的周期单位和精确 Todo 返回形状，无法可靠区分测试夹具错误与产品回归。
配置页读取默认重规划周期、管理器读取长 Todo 时，旧测试误报失败；本 PR 让精确选择返回单个 Todo，并继续核对长文本尾部、恢复条件和私有信息边界。
已验证 26 项相关测试通过；基础版本两处旧夹具失败，而两版本的隔离 File/SQLite 真实读取均保留完整尾部和非 owner 隐私边界。
本 PR 不改变运行时、周期配置、Todo 存储、用户操作步骤、权限或源码中已存在的精确读取行为。

改动思路

复用已有 cadence owner 和精确 Todo projection，修正当前 Chat 消费者的测试输入。实际读取先给有界 overview，随后按 selector 读完整 detail；测试必须分别表达这两个调用。没有必要添加新运行时 abstraction。

具体改动

• tests/test_chat_machine_configuration_api.py：默认单位改为 effective_turns，仍为 5。
• tests/test_chat_manager_details.py：假读取器按 selector 返回 list 或单个 todo；保留长 continuation 尾部、恢复条件和 scope 断言，并添加 missing/unavailable 的错误与隐私区分。两文件共 +39/-2。
  规范映射：docs/quota-allocation.md；revision d33cc13。settled effective Turn cadence、single exact Todo payload、bounded durable regression seam 均 implemented；另对照 canonical context projection 与 docs/reference/protocols/decision-scope-v0.md。已有 canonical/CLI 测试覆盖存储侧，本 PR 补的是 Chat 消费者边界。完整队列及 #5961 对照未发现同形低价值批量拆分。

对主干的风险

Head 的三个相关测试文件 26 passed；同一基础版本命令 22 passed / 2 failed，两失败均为旧夹具。独立隔离 File/SQLite + 实际 Node projection 在 B/H 都通过完整尾部、conditions/scopes、非 owner 隐私、missing/unavailable 和无写入检查。两次猜错文件名未执行测试，以及第一次 synthetic fixture 缺少 canonical schema_version，均保留为评审设置失败；后续正确来源的执行替代它们。diff-check、两文件 Ruff 通过；语义 advisory 不代表全树等价证明。未运行全仓、安装后 UI、live Lark 或远端 CI；测试维护不改变这些入口。未来重构检查：selector-aware local seam 已足够，无需新共享框架。

我的整体评价

APPROVE，无阻塞发现。这是恢复已有消费者回归信号的可独立维护改动，未以作者声明或测试数量代替 native readback。GitHub 当前调用账号与作者同为 loopx-agent，正式自批准受平台限制；本正文记录精确版本的批准结论，平台 COMMENTED 不等同于 GitHub APPROVED。未授权本轮合并。

English verdict: APPROVE — d971dc6; stale Chat fixtures repaired, 26 focused tests pass and independent File/SQLite base/head readbacks preserve full-tail/privacy behavior; no blocking findings.