feat: KOOK自动获取bot身份 + action/narration剥出气泡 + 渠道配置表单回填

agent-116 KOOK token即用+自动获取:
- kook/identity.go: Connect 调 GET /user/me 拿 bot 身份(bot_id/bot_username/bot_nickname), 失败容错
- 渠道列表 enrichAccountsWithBot 附 bot 身份; base_url空用默认/receive_mode默认ws(只填token可跑)

agent-117 action/narration 剥出气泡(主人需求: 动作/旁白不进气泡,按位置上/下):
- parseContent.splitBubble(content)->{body,head,tail}: body=纯文本(去标签), head=开头/中间action/narr(气泡上), tail=结尾(气泡下)
- renderContent 只渲染text(去内联灰斜体); chat.tsx BubbleNotes 泡外 ＊动作/旁白＊ 灰斜体小字(气泡外依附), SplitChatBubble 先剥splitBubble再splitParagraphs(body), <p>/\n\n/毛玻璃/SourceLabel不破坏
- 注: 非chat mode renderContent 现在丢弃action/narr(可接受)

Channels 配置表单回填(修save后输入框空):
- openConfig 从 data.accounts[0].config 回填 token(掩码提示)/base_url/平台字段/黑白名单/限流, 不再硬清空
全仓 build/vet/test绿+前端build绿

frontend/web/src/lib/parseContent.ts

@@ -1,8 +1,9 @@
 // Structured parser for persona action/narration XML tags (design/07 §1).
 // A single message `content` may mix plain text with `<action>…</action>` and
 // `<narration>…</narration>`; this splits it into an ordered parts array so the
-// UI can render inline (grey italic) while the structured data stays available
-// for downstream feeds (mind-map / state) via `summarizeContent`.
+// UI can position each block (inline, or as an attachment note via `splitBubble`)
+// while the structured data stays available for downstream feeds (mind-map /
+// state) via `summarizeContent`.
 
 export type ContentPartType = "text" | "action" | "narration";
 
@@ -64,3 +65,58 @@ export function summarizeContent(content: string): ContentSummary {
   }
   return summary;
 }
+
+/** Result of `splitBubble`: a renderable plain-text `body` plus the annotation
+ *  blocks that belong above (`head`) or below (`tail`) the bubble. `head`/`tail`
+ *  hold the leading / trailing action & narration parts in order; `body` is the
+ *  concatenated text with any annotation blocks removed. */
+export interface BubbleSplit {
+  body: string;
+  head: ContentPart[];
+  tail: ContentPart[];
+}
+
+/**
+ * Peel the `<action>` / `<narration>` blocks out of `content` so the bubble
+ * body is pure text while the annotations attach as notes outside it.
+ *
+ *  - `head` = the leading run of action/narration before the FIRST text part
+ *    (rendered above the bubble).
+ *  - `tail` = the trailing run of action/narration after the LAST text part
+ *    (rendered below the bubble).
+ *  - `body` = every text part concatenated and outer-trimmed.
+ *  - Annotation blocks interleaved BETWEEN text paragraphs have no stable text
+ *    anchor once the body collapses, so they lift to `head` — the only hard
+ *    requirement is that they never render inside the bubble.
+ *  - A content with no text part at all yields an empty `body` with everything
+ *    in `head`, so the notes alone carry the message.
+ */
+export function splitBubble(content: string): BubbleSplit {
+  const parts = parseContent(content);
+  let firstTextIdx = -1;
+  let lastTextIdx = -1;
+  parts.forEach((part, i) => {
+    if (part.type === "text") {
+      if (firstTextIdx < 0) firstTextIdx = i;
+      lastTextIdx = i;
+    }
+  });
+
+  const head: ContentPart[] = [];
+  const tail: ContentPart[] = [];
+  let bodyText = "";
+
+  parts.forEach((part, i) => {
+    if (part.type === "text") {
+      bodyText += part.text;
+      return;
+    }
+    if (firstTextIdx < 0 || i < firstTextIdx || (i > firstTextIdx && i < lastTextIdx)) {
+      head.push(part);
+    } else {
+      tail.push(part);
+    }
+  });
+
+  return { body: bodyText.trim(), head, tail };
+}