v0.1.29 · 视频声纹入库与 CI 回归修复

视频转写现在可以试听并选择指定说话人的原始语句，加入已有或新建人员的声纹档案；图文阅读滚动到底部会自动加载更多画面，并保留手动入口。

• 视频声纹样本使用任务已保存的转写时间线音频，修正视频起始偏移和延迟音轨对应问题；缺失历史音频返回 404，不改写历史任务或样本。
• 修复图文时间按钮定位。阅读画面按 20 个一批自动加载，逐句明细继续支持滚动加载。
• 修复浏览器夹具刷新时的 capabilities/system 断连、摘要轮询引起的阅读位置重置、取消测试计时和截图编辑后的就绪判断。新增桌面/390 px 加载、错误恢复及延迟画面回归，保留原有超时和零整套测试重试。

From v0.1.26, this release adds video speaker preview/selection and voiceprint enrollment, progressive scene loading, and timeline-aligned sample extraction. Browser fixtures and readiness regressions cover desktop/mobile recovery without weakening deadlines or adding whole-test retries.

从正式发布的 v0.1.26 或已部署的 v0.1.27/v0.1.28 源码升级：数据迁移：无；无需额外升级备份。历史数据库、媒体和声纹样本不改写，模型、依赖、推理参数和浏览器存储生命周期不变。重建前端，并刷新原运行的 API、ASR/TTS supervisor 和 executor。更早版本须覆盖中间版本迁移要求。

Upgrading from released v0.1.26 or deployed v0.1.27/v0.1.28 source requires no data migration or additional upgrade backup. Rebuild the frontend and refresh all previously running service components. Earlier upgrades must cover intervening migrations.

升级说明 / Upgrade guide · CI 修复经验

v0.1.27 和 v0.1.28 是 Windows 检查失败后保留的未发布审计标签，不移动或复用。This release delivers their features and fixes; both failed tags remain immutable and unpublished.

提交 / Commit: a851e5abd775ba71c94e797c3296ad36cb817764

验证 / Validation:

• main · Native Windows smoke: passed.
• main · Linux quality gates: passed.
• tag · Linux quality gates: passed.
• tag · Native Windows smoke: passed.

Both tag workflows passed 149 browser tests, with 5 optional external Office sample skips. Linux backend: 407 passed / 16 skipped; native Windows backend: 395 passed / 28 skipped. CI exercises mock inference and native service lifecycle. Local patch checks: 40 repeated desktop/mobile cases, clean-source build and 31 documentation/API/version checks, plus temporary-tag version contracts.