fix(quota): align deferred priority with action selection projection

[huangruiteng]

Summary

• Reuse the existing ready-deferred priority frontier when building the quota action portfolio, so a P1 candidate that cannot run is not presented as an eligible alternative.
• Feed the same preemption into the TypeScript-owned explicit-selection qualifier. The result is a typed deferred response instead of a later selection conflict; no false receipt is committed.
• Preserve independent P1 fallback when the P0 deferred resume condition is still pending, and preserve equal-priority P0 selection.

Validation

• uv run --extra test python -m pytest -q tests/control_plane/test_quota_settlement_cli.py (70 passed)
• uv run --extra test python -m pytest -q tests/control_plane/test_quota_action_portfolio.py tests/control_plane/test_quota_selection.py tests/control_plane/test_agent_scope_frontier_contract.py (34 passed)
• Focused new regressions (2 passed), Ruff, py_compile, git diff --check.

Product path and scope

CLI and Codex App managed quota selection use the corrected shared projection. No frontend or Lark UI change: neither has a separate action-portfolio selection source. No private Goal state or credentials are included. The preexisting deferred-resume priority rule is reused, not broadened to a blanket P0-versus-P1 prohibition.

Residual risk

This is a selection-projection alignment slice. It does not change the deferred lifecycle policy or receipt replay semantics. Control-plane owner review is required before merge.

[huangruiteng]

Approval conclusion (author-owned PR; GitHub blocks formal self-approval)
Reviewed exact head: af65f59

动机

quota should-run 曾把一个 P1 列成可选替代项，但同一批权威 Todo 里已有更高优先级、恢复条件已满足的 deferred successor。显式绑定 P1 时，资格判定先说 qualified，最终 frontier 又选中 deferred 项，形成冲突。这个 PR 要完成的边界是“推荐与绑定一致”，不是重写 P0/P1 调度政策。

改动思路

权威事实仍来自当前 Goal Todo summary；复用现有 _selected_candidate_priority_frontier 判定，而非新建优先级表。候选投影和显式选择两个入口都消费这条判定，后者仍交给 TS qualifyActionSelection 产生 typed deferred。只改一边会留下另一边的矛盾；不发货则继续误导 agent。没有新增持久状态、schema 或交易权限。

具体改动

关键代码讲解

• quota_runnable_action_candidates：从 capability-gated open advancement 候选构建 portfolio 时，排除会被已就绪高优先级 deferred successor 抢占的候选；同级和未就绪条件不受此规则排除。
• _delivery_preemptions_for_route：仅在没有更高优先级的既有硬门禁时，把相同 frontier 结果传给 TS 资格判定，显式 P1 因而返回 deferred，而不是先 qualified 再发生 selection conflict。
• CLI 回归测试覆盖 portfolio 不再推荐 P1、P1 的未提交回执、同轮同级 P0 可绑定，以及未来才恢复的 P0 不阻止独立 P1。

对主干的风险

主要风险是 hot path 多一次现有 Python frontier 调用，以及 Todo summary 若不完整时与最终 frontier 同源的漏判。没有第二套阈值；后续 TS 整笔事务迁移应整体迁走此既有判定，而非在 TS 复制一份。针对最强反例，未就绪 P0 + 独立 P1 的真实 CLI 路径仍通过；未引入跨 agent 授权或 receipt replay 改动。70 个 quota CLI、34 个相邻控制面、28 个下游选择/投影测试通过，Ruff、py_compile 和 diff 检查通过；本 Goal 的 review 配置为不等待/查询远端 CI。

语义与 CI 对齐

现有 quota 契约已规定“严格更高优先级且已就绪的 deferred successor 先做生命周期裁决”，并允许等待中的高优先级项让独立低优先级工作继续。本改动沿用该词汇与决策 owner；无新 schema。合并后仍需在安装版对一个实际 Goal 读回，本地合成 fixture 不冒充生产验收。

我的整体评价

没有发现阻断性问题。这个切片对所报投影矛盾是完整且可回退的修复，保留 agent 对一般 runnable P0/P1 的选择，不实施一刀切禁用 P1。评审结论为可接受；但它改变控制面行为，按仓库规则必须交维护者合并，作者不自合并。

English verdict: APPROVE - exact head af65f59 reuses the existing ready-deferred frontier across portfolio and TS-owned explicit selection; local CLI and downstream validation passed, with post-merge installed readback still required.