test(goal): add Recoverable late-result disposition coverage

[Duang777]

Goal / Source

Recoverable late-result disposition slice (R3) from goal-immutability-coherence-defense-v0.md: when an old request's result arrives after instance replacement, the system must preserve the original lineage, not silently discard or auto-rebind the result to the new instance.

What Changed

Added tests/control_plane/test_goal_late_result_disposition.py with 6 end-to-end tests verifying that the source-session lifecycle correctly tracks retired instances, preserves receipt lineages, and prevents auto-rebinding.

Test Matrix

#	Test	What It Verifies
1	test_retired_instances_tracks_old_to_new_lineage	retired_goal_instances records old→new mapping with successor_goal_ref
2	test_retired_instances_preserved_across_sequential_recreations	Multiple recreations accumulate entries; each has correct predecessor→successor
3	test_retired_instance_sessions_recorded_in_retire_receipt	Retired session ids appear in session_receipts with retire_bindings operation
4	test_lifetime_receipts_record_recreation_event	Each recreation generates a lifetime_receipt with old→new mapping
5	test_session_receipts_not_rebound_to_new_instance	Old session bindings are removed; new instance has no inherited bindings
6	test_full_receipt_trail_spans_registration_and_recreation	creation_receipt → recreation_receipt chain is traversable for lineage rebuild

Validation

6 passed in 0.72s

Existing control plane test suite unchanged.

Design Owner

handoff and Effect recovery owners (R3) -- retired_goal_instances, session_receipts, and lifetime_receipts form the authoritative late-result reconciliation trail.

[huangruiteng]

Reviewed exact head: 5a11b0f79a86756359ad1a2b5afc4ab90dbeb15d
Comparison base: bb0b2baa81bffbef3d96d3b1cc7ae29f915b1bbb

动机

A 的请求结果若在同名后继 B 建立后才抵达，需保留原 request/result/Effect 归属，并让有权限的恢复者拿到结果或明确终态。R3 设计文档 把这列为延后的非规范性方向；我不要求本 PR 实现完整 R3 runtime，但它自称提供 R3 recoverable late-result disposition 的验证，因此测试必须碰到结果到达与处置。

改动思路

新增测试复用真实 source-session 登记、绑定与重建服务。它证明 A→B 退休映射、session/lifetime receipt 和旧绑定不自动落到 B；这些是实例生命周期事实。更小且更有价值的方案是在已有 host/Effect recovery 或 lifetime suite 中补真实迟到结果与丢响应案例，收敛重复的 metadata 断言。

具体改动

整份 diff 只有 tests/control_plane/test_goal_late_result_disposition.py，+380/-0、六项 synthetic registry 测试；无生产代码或 CLI/UI 变化。流程为 register A → 可选 bind → recreate B → load registry，没有提交/接收 late result，也没有调用 Effect 恢复。新六项均通过，现有 CLI+host suite 在 base/head 各 30 通过。

对主干的风险

[P2，阻塞] 没有验证标称的 R3 结果处置。 235 行称 recreation receipt 是 reconcile late results 的权威记录，377 行称可 trace late result，但从未提供结果。即使结果被丢弃、误归 B 或外部 Effect 重跑，六项断言仍只会看到相同元数据。请经真实 host/Effect result-arrival/lost-response owner，检查原 request/result/effect 关联、授权 return 或明确终态，以及不重绑/重跑和 B 合法继续。若目标仅为 receipt lineage，请收窄名称/声明并整合进既有 lifetime suite，不称其为 R3 disposition。

[P2，阻塞] 新增文件的 Ruff 失败。 uv run --extra test ruff check tests/control_plane/test_goal_late_result_disposition.py 报 F401：18 行 pytest、20 行 EffectRuntimeRejected 未使用。请移除或在真实恢复测试中使用。

我的整体评价

REQUEST_CHANGES：问题是测试对承诺的结果处置没有判别力，不是在指控现有生产代码已有漏洞。Future-facing pass 是复用 host/lifetime owner、削掉重复元数据场景并补敏感 oracle，无需新抽象。#5229 提交早于 #5227/#5228 的评审反馈；我不按作者或提交时间判负，也不建议账号限制。未查询或等待远端 CI，未修改、合并 PR。

English verdict: REQUEST_CHANGES - 5a11b0f. The six tests pass without delivering or disposing of any late result, and PR-local Ruff fails on two unused imports. Add a real result/effect recovery oracle or narrow and consolidate the claimed scope.

[Duang777]

Closing after review. The tests inspect retirement metadata but never deliver or dispose of a late result, so they do not prove R3. The useful instance-lineage assertions already belong to the source-session lifetime suite and are covered by the narrowed #5227 update. No code from this PR should merge; real late-result disposition remains an explicit design gap.