fix(asr): keep recordings when streaming recognition fails or stalls

- Keep recording after a streaming ASR error and retry the captured PCM
  through batch ASR before reporting a failure.
- Treat an expired ASR wait as a stall so batch fallback still runs.
- Prime the managed Volcengine stream on mic touch and adopt it only while
  fresh and open.
- Auto-scroll the live transcript to the newest recognized text.

CHANGELOG.md

@@ -13,6 +13,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - **Host-only analytics recording**: the keyboard extension no longer opens the analytics or keyboard-usage SQLite queues. The host app is the only process that records and uploads events, so hiding the keyboard cannot hold WAL locks and get killed as `0xdead10cc`. Keyboard-activated and per-keystroke usage events are no longer collected. / **仅主 App 记埋点**：键盘扩展不再打开统计与键盘用量 SQLite 队列；开库、记账与上传只在主 App 进行，收起键盘时不会因 WAL 锁被系统以 `0xdead10cc` 终止。不再采集「键盘激活」和逐次按键用量。
 
 ### Fixed
+- **Faster managed dictation feedback**: begin the managed Volcengine streaming connection while the microphone touch primes audio, reuse it when recording starts, and keep the live transcript automatically scrolled to the newest recognition text. The early connection is reused only while it is fresh (under 10 seconds) and still open; otherwise it is closed and a new one opens, so an unused early connection can no longer make the next dictation fail. / **托管听写反馈提速**：在麦克风触摸预热音频时同步建立托管火山流式连接，录音开始后直接复用，并让实时转写自动滚动到最新识别文字。提前建立的连接仅在新鲜（10 秒内）且仍可用时复用，否则关闭并重新建立，未被使用的提前连接不会再导致下一次听写失败。
+- **Recording kept when live recognition fails**: a streaming recognition error while you were still speaking ended the dictation immediately and discarded the audio. Recording now continues, and when you finish, the recorded audio is retried with batch recognition before any error is shown. After a cancelled or failed dictation, that batch retry also stopped working for the rest of the session; it now works again from the next dictation. / **实时识别失败时保留录音**：说话过程中流式识别一旦出错，本次听写会立即结束并丢弃录音。现在会继续录音，说完后先用批量识别重试已录下的音频，仍失败才提示错误。此前一次听写被取消或失败后，本次会话内的批量重试也会失效；现在从下一次听写起恢复正常。
+- **Stalled cloud ASR losing the recording**: when the streaming recognizer went quiet and the result wait expired, the utterance ended as "no speech" even though the audio was captured. A stalled recognizer now counts as a failure, so the batch ASR fallback retries the recorded audio before giving up. / **云端识别卡住时丢失录音**：流式识别无响应、等待结果超时后，即便录音已采集，本次语音也会直接报「未检测到语音」。现在识别卡住按失败处理，会先用批量识别重试已录下的音频再放弃。
 - **Speak as Me output shape**: the skill could answer with raw JSON variants or echo the personal-style XML tags into the result. The prompt now requires one ready-to-send plain-text rewrite, and length follows your natural voice instead of a mechanical ±20% cap. / **「像我所说」输出形态**：该技能可能返回原始 JSON 候选，或把个人风格的 XML 标记带进结果。现在提示词要求只输出一段可直接发送的文本，长度改为跟随你的自然表达节奏，不再机械限制 ±20%。
 - **Clipboard history reverting deletes**: history is one shared blob, and the main App kept an in-memory copy that only refreshed when a screen appeared — so a delete made there could write back entries the keyboard had just captured or removed, and a clear done in the keyboard could be undone entirely. Both processes now announce their writes to each other, and the main App re-reads history whenever it returns to the foreground. / **剪贴板历史撤销删除**：历史记录是整块共享数据，而主 App 仅在页面出现时刷新内存副本——因此在主 App 中删除一条，可能把键盘刚采集或刚删除的条目重新写回，在键盘里执行的清空也可能被整体撤销。现在两端写入时会互相通知，主 App 每次回到前台也会重新读取历史。
 - **Keyboard crash loop on memory relief**: recording a memory milestone could request relief while relief was already running, and the two sides called each other until the keyboard extension exhausted its stack (SIGSEGV grouped under `KeyboardViewController.viewDidLoad`). Reentrancy guards on both the controller and the telemetry side break the cycle. / **键盘内存降级递归崩溃**：记录内存里程碑时可能再次请求降级，两侧互相回调直至扩展栈溢出（崩溃聚合在 `KeyboardViewController.viewDidLoad`）。控制器与遥测两侧均加重入保护，打断循环。