ci: preserve Windows test crash diagnostics

[yimi528]

Summary

• keep the Windows one-retry mitigation observable
• propagate Node fatal-error and uncaught-exception report flags to per-file test subprocesses
• preserve per-attempt npm test logs
• upload Windows Node reports and logs when the job fails or only passes after retry

Context

Addresses the diagnostic follow-up discussed in #413. This complements the one-retry mitigation in #410; it does not change test exit codes or treat a retry as a root-cause fix.

Validation

• YAML parsed successfully
• PowerShell diagnostic and retry blocks passed parser checks
• Node 24 diagnostic smoke test produced a report while preserving a non-zero exit code
• retry pipeline preserved both success and failure exit codes
• git diff --check passed

Follow-up

The next Windows failure should provide a Node diagnostic artifact so a separate root-cause fix can target the observed async teardown or native crash rather than guessing.

[x-lixu]

感谢补充 Windows CI 的日志和诊断方案。当前改动未修复已确认的根因，且重试条件覆盖全部测试失败，而不只限于已识别的静默退出，可能掩盖间歇性的真实缺陷。维护方目前优先处理可验证的具体问题，暂不引入通用重试和额外诊断流程，因此先关闭；后续若有明确复现和根因，欢迎提交针对性修复。