[P2] 验收作业的分步预算之和 80 分钟超过作业上限 75 分钟，而最常见的改动就会走到这条路径

把 OceanBase 启动拆成独立步骤后，本作业三步上限是 15（oceanbase_setup）+ 50（native_acceptance）+ 15（acceptance_scenarios）= 80 分钟，而 timeout-minutes 仍是 75，本行注释又写"保持既有总预算"。base 里数据库启动在同一个 50 分钟步骤内执行，合计 50 + 15 = 65，还给证据扫描和 teardown 留了 10 分钟。

影响：oceanbase_acceptance 已加入 [shared.runtime].suites，任何 src/**、openapi/**、pyproject.toml、uv.lock 改动都会同时选中两个范围，也就是最常见的一类 PR。当 OceanBase 起得慢（up --wait-timeout 660 已允许 11 分钟，参数轮询最多再 60 秒）、原生用例又接近 50 分钟上限时，作业会在 75 分钟处被整体取消，Require all selected acceptance work and evidence 来不及执行，失败只表现为 job 超时，看不出是哪一段验收没跑。

建议：让三步之和不超作业上限（例如 native_acceptance 降为 45 分钟，或上限提到 85 并同步改注释），并把 teardown 与证据上传的时间也计入；tests/test_e2e_workflow_contract.py 可以顺手断言"分步上限之和 ≤ 作业上限"，避免以后再漂。

Fixed in 342cf9ad. The OceanBase job now allows 90 minutes: 15 for startup + 50 for native acceptance + 15 for replay, with 10 minutes reserved for diagnostics, evidence and cleanup. A regression check covers combined, native-only and replay-only budgets.

I also integrated #2515 at 08a34940 into this branch and fixed the combined workflow: the container ID is passed through a setup output, the native deadline starts in its own step, and JUnit validation/upload use the same file. Changes to the shared OceanBase helper now select native acceptance as well. Pre-teardown diagnostics are preserved.

The regression checks fail against the uncorrected combination and pass with the fix; local checks and workflow lint pass. Real OceanBase acceptance still requires the new CI run. This PR now depends on #2515, which should be merged first; the #2515 branch was not changed.

[已修复] 分步预算改为 15(setup) + 50(native) + 2(diagnostics) + 15(replay) = 82 ≤ 作业上限 90（原 15+50+15=80 对 75）。新增的 oceanbase_setup 步骤内部也自洽：timeout 700s compose up --wait 加 oceanbase_ci.py wait --timeout 90 共 790s < 步骤上限 900s；native 步骤 native_deadline=$((SECONDS + 2850))（47.5 min）< 步骤上限 50 min。

补充证据（exact head 105e7a94，WSL venv 3.13.11）：tests/test_oceanbase_ci.py + tests/test_ci_changes.py + tests/test_e2e_workflow_contract.py 共 178 passed。

验证缺口（不构成新 Finding，但影响合并判断）：该 head 的 Main run 42390978248 结论是 cancelled，其中 tests (3.12, e2e) 被取消；suite 作业的 if 含 contains(fromJSON('["success","skipped"]'), needs.tests.result)，于是 CI preflight 明明输出全 true（含 harness / oceanbase_acceptance / upgrade），仍有 30 个 suite 作业 skipped，Acceptance (oceanbase) 连 check 都没有产生。因此 82/90 的预算与本 PR 新增的 setup 路径目前没有一次完整绿跑作为证据。

另核对 scripts/oceanbase_ci.py 的解析列位：values = {row[4]: row[6] ...} 与 e2e-harness.yml 中 awk '$5 == "datafile_size" && $7 == "4G"' 一致，对应 SHOW PARAMETERS 的 name / value 列（4.x 输出为 zone, svr_type, svr_ip, svr_port, name, data_type, value, …）。测试 fixture 的 7 列形状与之一致，不是测试迎合实现造成的错位。

Thank you for confirming the corrected 82/90-minute budget. The latest master conflict resolution and CI fixes are pushed in b95e6a5. Both branches include master 7fef43d, pass make check, and combine locally without conflicts. The scope-specific workflow selection remains on this branch.

Native authorization scenarios reuse only the module-scoped database schema, while retaining separate Scopes, idempotency keys and Server lifecycles. Migration-recovery scenarios keep isolated databases. The resource producer also has a checksum-preserving official Release API fallback for the observed HTTP 504 download failures.

The previous run passed Python 3.11–3.13, but its 3.14 unit runner lost communication with GitHub. A failed-only rerun passed that unit job; remaining validation was superseded by this conflict-resolution push. These results do not close the native validation gap. Run 42573616986 validates the newly pushed head; I will report the actual backend outcome before claiming acceptance is complete.

The same bounded, checksum-verified downloader now covers skill-up and pre-staged Windows NSIS tools. The plugin’s ordinary release URL reproduced HTTP 504 locally, while the official asset API returned bytes matching both the pinned SHA256 and Tauri’s upstream SHA1. This verifies the alternate artifact route; the native Windows installer and backend suites still require the linked CI run.