feat(serial): Step 3 读线程生命周期修复

[Gyanano]

RFC #3 Step 3：读线程生命周期

跟踪 issue：#3（前置：#4、#5、#6）

改动

后端

• disconnect() 返回读线程 JoinHandle，bounded join（500ms，is_finished 轮询）移到 Tauri 命令层、大锁外执行，取代原锁内 200ms sleep；join 超时则 detach 不阻塞 UI
• connect() 开头防御性 join 残留读线程 → close→reopen EBUSY 竞态根治
• 读线程致命错误不再静默：错误写入共享状态 → 下次状态轮询（1s）is_connected=false 且 ConnectionStatus 新增 connection_error 字段；惰性清理写句柄与录制
• 读线程任何退出路径（shutdown / 致命错误）都会 FrameSegmenter::flush() 打捞残留分帧，尾部数据不再蒸发

前端

• 状态轮询检测到「连接意外断开」转换时 toast 告警（中英 i18n），手动断开不打扰
• 修复定时发送僵尸态：断开连接或清空内容时开关自动复位（原需手动关开才能恢复）

验证

• cargo test 34/34 全绿，零警告；npm run build 通过
• 测试调整：golden_reader_death_is_silent_today（钉住旧坏行为）→ reader_death_marks_connection_lost_and_salvages_pending（钉住新行为）；新增 reconnect 竞态测试与 flush 单测
• 有头应用人工硬件验证（FTDI 环回）：
  • 拔线 → 1s 内离线 + toast + 定时开关自动复位 ✅
  • 拔线重插立即重连成功（原 EBUSY 竞态）✅
  • 快速断连/重连循环 ✅
  • 断开时数据完整保留 ✅
  • 收发/录制/导出回归 ✅

范围外

• 前端日志拉取仍为轮询 → Step 4 事件推送
• 高负载接收丢字节（轮询克隆饿死读线程）→ bug(serial): 高负载持续接收时驱动层疑似丢字节（读线程被轮询克隆饿死） #9
• TX 队列 → feat(serial): TX write-behind 发送队列 + 背压与失败上报 #7

---

Note

Medium Risk
Changes core serial connection lifecycle (thread join timing, lazy disconnect on reader death, and reconnect races); mistakes could block polling or leave ports stuck, though bounded join and tests mitigate this.

Overview
Implements RFC #3 Step 3 so serial read-thread shutdown, fatal errors, and reconnect behave correctly instead of leaving a “connected” UI or dropping tail bytes.

Backend: disconnect() now returns the reader JoinHandle and does fast cleanup under the manager lock; bounded join (500ms) runs in the Tauri disconnect_port command after releasing the lock (replacing a lock-held 200ms sleep). connect() joins any stale reader before reopen to avoid EBUSY on quick replug. Fatal read errors are stored on the reader thread; get_status lazily marks the session disconnected, exposes new connection_error on ConnectionStatus, and tears down the port/recording. On any reader exit, FrameSegmenter::flush() emits buffered RX so partial frames are not lost.

Frontend: Status polling shows a one-time toast when transitioning from connected to disconnected with connection_error (manual disconnect has no error). SendPanel turns off scheduled sending when disconnected or the payload is cleared, fixing a “scheduled active” zombie state. EN/zh-CN strings added for connection loss.

^{Reviewed by Cursor Bugbot for commit 4cf035c. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: serverGenReqId_8937df3d-8a14-48a5-a8d2-87356143fe6f)