fix: preserve quote context (ref_msg) for voice messages

[gaoyangz77]

Summary

• Extract shared ref_msg handling from the TEXT branch of bodyFromItemList into a reusable applyRefMsg helper
• Apply the same helper to the VOICE branch so voice messages with STT text preserve quoted context

Previously, when a user sent a voice message quoting a previous message, the agent only received the STT transcription with no indication of the quoted message. Now the output matches the TEXT behavior: [引用: <quoted content>]\n<voice text>.

Closes #48

Changes

src/messaging/inbound.ts

Before:

// TEXT: handles ref_msg ✅
if (item.type === TEXT && item.text_item?.text != null) {
  const text = String(item.text_item.text);
  const ref = item.ref_msg;
  // ... ref_msg logic ...
}
// VOICE: ignores ref_msg ❌
if (item.type === VOICE && item.voice_item?.text) {
  return item.voice_item.text;
}

After:

function applyRefMsg(text: string, ref: MessageItem["ref_msg"]): string {
  // shared ref_msg logic
}

// TEXT: ✅
if (item.type === TEXT && ...) {
  return applyRefMsg(String(item.text_item.text), item.ref_msg);
}
// VOICE: ✅
if (item.type === VOICE && ...) {
  return applyRefMsg(item.voice_item.text, item.ref_msg);
}

Test plan

• Send a text message, then quote it with a voice message — agent should see [引用: ...] prefix
• Send a voice message without quoting — agent should see only the STT text (no regression)
• Quote a media message (image/video) with a voice message — agent should see only the STT text (media quotes are intentionally excluded)

[gaoyangz77]

Friendly ping / 轻轻 bump 一下～

这个 PR 修复了语音消息引用时上下文不正确的问题，能让引用内容更准确，减少聊天场景里的理解偏差；同时也 closes #48。

This PR fixes the incorrect context used in voice message quoting, which should make quoted voice replies more accurate and reduce confusion in chat flows; it also closes #48.

CI is already passing, so it should be ready for review when convenient. Thanks!

[gaoyangz77]

Gentle ping / 再轻轻 bump 一下～

This PR is still ready for review when convenient. CI has been passing, and the change is scoped to fixing quote context (ref_msg) for voice messages so quoted voice replies use the correct context.

有空的时候麻烦帮看一下，感谢！