fix: only emit fork chip at actual divergence point

Changed last_fork.clone() to last_fork.take() in render_tree_path
for both assistant and user branches. This prevents fork_info from
propagating to every downstream message, ensuring the chip appears
only at the actual divergence point in the conversation tree.

src/ai/insight_chat.rs

@@ -2511,8 +2511,10 @@ fn render_tree_path(
         std::collections::VecDeque::new();
     let mut node_ids: Vec<u64> = Vec::new();
     let mut fork_info: Vec<Option<ForkInfo>> = Vec::new();
-    // Nearest divergence seen so far, carried forward across nodes (rendered or
-    // not) so every message at/after a fork shows the indicator.
+    // Nearest divergence seen so far, carried forward across non-rendered
+    // nodes (e.g. tool-dispatch) so the fork is attributed to the next
+    // rendered message. Cleared after the first rendered message consumes it,
+    // so the chip appears only at the actual divergence point.
     let mut last_fork: Option<ForkInfo> = None;
 
     for node in path {
@@ -2571,7 +2573,7 @@ fn render_tree_path(
                     tools,
                 });
                 node_ids.push(node.id);
-                fork_info.push(last_fork.clone());
+                fork_info.push(last_fork.take());
             }
               "user" => {
                 let is_initial = user_turns_seen == 0;
@@ -2585,7 +2587,7 @@ fn render_tree_path(
                     tools: Vec::new(),
                 });
                 node_ids.push(node.id);
-                fork_info.push(last_fork.clone());
+                fork_info.push(last_fork.take());
             }
             _ => continue,
         }