test: 修正并行完成顺序断言

[Ev3rGan]

关联：#1、#3（Relates to #1 and Relates to #3）

背景

Parent #1 标准化 PR 合并后的 main push CI run 33513019062 提供了明确的红灯证据：tests/test_cli.py::test_mixed_batch_cli_exposes_both_modes_and_ordered_results 把 parallel Tool 的完成事件硬编码为 ToolCall 原始顺序，但实际合法完成顺序为 read-b、read-a，导致 Linux CI 失败。

修复的测试契约

• parallel tool_execution_end 事件反映真实 completion order，因此可以是 b-before-a；
• 模型可见的 ToolBatchResult.results 必须稳定保持原始 ToolCall order；
• 测试使用确定性 barrier 强制 b-before-a，同时分别证明上述两个事实；
• sequential batch 的事件与结果顺序继续按调用顺序；
• 不修改生产代码，不增加 skip/ignore，不缩窄 CI，也不掩盖并行行为。

精确身份

• Base：fc85729b6bd10c84710ed4bb7686fc48158ae002
• Head：e3cf13ed825809ee9225f93a1891525fc6ab4e6e
• 1 个提交，2 个路径
• Diffstat：34 additions / 10 deletions
• Stable patch-id：b84a9d1a92cf9dbe5bc4578090dc6e1d2331b87b

路径仅为：

• tests/test_cli.py
• tests/test_tool_runtime.py

验证

Windows：

• focused suite 连续 30 次通过
• full pytest 两次：334 passed，1 个 network-contract skip
• Ruff lint/format、strict mypy：PASS

官方 python:3.11-slim Linux：

• focused suite 连续 20 次通过
• full pytest：331 passed，4 个预期 platform/network skips
• Ruff lint/format、strict mypy：PASS

此外，build、git diff --check、Standards review 与 Issue #3/CI Spec review 均通过且 0 findings。未读取环境凭据，未调用 Provider 或 Harness。

本 PR 不自动关闭 #1 或 #3；合并必须等待 exact head 的 GitHub Actions SUCCESS。合并后的 main SHA 还必须再次通过 push CI，之后才能恢复被冻结的仓库配置阶段。

---

English appendix

Relates to #1 and Relates to #3. Main push CI run 33513019062 exposed an incorrect test assumption: parallel completion events are allowed to arrive in completion order, while the model-visible ToolBatchResult.results must remain in original ToolCall order.

This test-only hotfix uses a deterministic barrier to force b-before-a and proves both contracts independently. It changes no production code, adds no skip/ignore, and does not narrow CI.

Exact identity: base fc85729b6bd10c84710ed4bb7686fc48158ae002, head e3cf13ed825809ee9225f93a1891525fc6ab4e6e, one commit, two files, 34 additions / 10 deletions, stable patch-id b84a9d1a92cf9dbe5bc4578090dc6e1d2331b87b.

The same tree passed repeated focused and full suites on Windows and the official Python 3.11 slim Linux image, plus Ruff, strict mypy, build, diff checks, and fixed-point Standards/Spec reviews. Merge is gated on successful CI for both the exact PR head and the resulting main merge SHA.