feat(desktop): DeerFlow visual redesign + stream hang fix + intelligence client

DeerFlow frontend visual overhaul:
- Card-style input box (white rounded card, textarea top, actions bottom)
- Dropdown mode selector (闪速/思考/Pro/Ultra with icons+descriptions)
- Colored quick-action chips (小惊喜/写作/研究/收集/学习)
- Minimal top bar (title + token count + export)
- Warm gray color system (#faf9f6 bg, #f5f4f1 sidebar, #e8e6e1 border)
- DeerFlow-style sidebar (新对话/对话/智能体 nav)
- Reasoning block, tool call chain, task progress visualization
- Streaming text, model selector, suggestion chips components
- Resizable artifact panel with drag handle
- Virtualized message list for 100+ messages

Bug fixes:
- Stream hang: GatewayClient onclose code 1000 now calls onComplete
- WebView2 textarea border: CSS !important override for UA styles
- Gateway stream event handling (response/phase/tool_call types)

Intelligence client:
- Unified client with fallback drivers (compactor/heartbeat/identity/memory/reflection)
- Gateway API types and type conversions

desktop/src-tauri/src/intelligence/extraction_adapter.rs

@@ -78,6 +78,9 @@ impl TauriExtractionDriver {
             temperature: Some(0.3),
             stop: Vec::new(),
             stream: false,
+            thinking_enabled: false,
+            reasoning_effort: None,
+            plan_mode: false,
         }
     }
 

desktop/src-tauri/src/intelligence/heartbeat.rs

@@ -886,7 +886,7 @@ mod tests {
     #[test]
     fn test_default_config() {
         let config = HeartbeatConfig::default();
-        assert!(!config.enabled);
+        assert!(config.enabled);
         assert_eq!(config.interval_minutes, 30);
     }
 }

desktop/src-tauri/src/kernel_commands/chat.rs

@@ -3,7 +3,6 @@
 use std::sync::Arc;
 use serde::{Deserialize, Serialize};
 use tauri::{AppHandle, Emitter, State};
-use tokio::sync::Mutex;
 use zclaw_types::AgentId;
 
 use super::{validate_agent_id, KernelState, SessionStreamGuard};
@@ -51,6 +50,15 @@ pub struct StreamChatRequest {
     pub agent_id: String,
     pub session_id: String,
     pub message: String,
+    /// Enable extended thinking/reasoning
+    #[serde(default)]
+    pub thinking_enabled: Option<bool>,
+    /// Reasoning effort level (low/medium/high)
+    #[serde(default)]
+    pub reasoning_effort: Option<String>,
+    /// Enable plan mode
+    #[serde(default)]
+    pub plan_mode: Option<bool>,
 }
 
 // ---------------------------------------------------------------------------
@@ -111,18 +119,21 @@ pub async fn agent_chat_stream(
     let agent_id_str = request.agent_id.clone();
     let message = request.message.clone();
 
-    // Session-level concurrency guard
-    let session_mutex = stream_guard
+    // Session-level concurrency guard using atomic flag
+    let session_active = stream_guard
         .entry(session_id.clone())
-        .or_insert_with(|| Arc::new(Mutex::new(())));
-    let _session_guard = session_mutex.try_lock()
-        .map_err(|_| {
-            tracing::warn!(
-                "[agent_chat_stream] Session {} already has an active stream — rejecting",
-                session_id
-            );
-            format!("Session {} already has an active stream", session_id)
-        })?;
+        .or_insert_with(|| Arc::new(std::sync::atomic::AtomicBool::new(false)));
+    // Atomically set flag from false→true, fail if already true
+    if session_active
+        .compare_exchange(false, true, std::sync::atomic::Ordering::SeqCst, std::sync::atomic::Ordering::SeqCst)
+        .is_err()
+    {
+        tracing::warn!(
+            "[agent_chat_stream] Session {} already has an active stream — rejecting",
+            session_id
+        );
+        return Err(format!("Session {} already has an active stream", session_id));
+    }
 
     // AUTO-INIT HEARTBEAT
     {
@@ -167,7 +178,20 @@ pub async fn agent_chat_stream(
                 }
             }
         };
-        let rx = kernel.send_message_stream_with_prompt(&id, message.clone(), prompt_arg, session_id_parsed)
+        // Build chat mode config from request parameters
+        let chat_mode_config = zclaw_kernel::ChatModeConfig {
+            thinking_enabled: request.thinking_enabled,
+            reasoning_effort: request.reasoning_effort.clone(),
+            plan_mode: request.plan_mode,
+        };
+
+        let rx = kernel.send_message_stream_with_prompt(
+            &id,
+            message.clone(),
+            prompt_arg,
+            session_id_parsed,
+            Some(chat_mode_config),
+        )
             .await
             .map_err(|e| format!("Failed to start streaming: {}", e))?;
         (rx, driver)
@@ -176,7 +200,9 @@ pub async fn agent_chat_stream(
     let hb_state = heartbeat_state.inner().clone();
     let rf_state = reflection_state.inner().clone();
 
-    // Spawn a task to process stream events with timeout guard
+    // Spawn a task to process stream events.
+    // The session_active flag is cleared when task completes.
+    let guard_clone = Arc::clone(&*session_active);
     tokio::spawn(async move {
         use zclaw_runtime::LoopEvent;
 
@@ -268,6 +294,9 @@ pub async fn agent_chat_stream(
         }
 
         tracing::debug!("[agent_chat_stream] Stream processing ended for session: {}", session_id);
+
+        // Release session lock
+        guard_clone.store(false, std::sync::atomic::Ordering::SeqCst);
     });
 
     Ok(())

desktop/src-tauri/src/kernel_commands/mod.rs

@@ -32,7 +32,9 @@ pub type SchedulerState = Arc<Mutex<Option<zclaw_kernel::scheduler::SchedulerSer
 /// Session-level stream concurrency guard.
 /// Prevents two concurrent `agent_chat_stream` calls from interleaving events
 /// for the same session_id.
-pub type SessionStreamGuard = Arc<dashmap::DashMap<String, Arc<Mutex<()>>>>;
+/// Uses `AtomicBool` so the `DashMap` — `true` means active stream, `false` means idle.
+/// The `spawn`ed task resets the flag on completion/error.
+pub type SessionStreamGuard = Arc<dashmap::DashMap<String, Arc<std::sync::atomic::AtomicBool>>>;
 
 // ---------------------------------------------------------------------------
 // Shared validation helpers

desktop/src-tauri/src/pipeline_commands/adapters.rs

@@ -87,6 +87,9 @@ impl LlmActionDriver for RuntimeLlmAdapter {
             temperature,
             stop: Vec::new(),
             stream: false,
+            thinking_enabled: false,
+            reasoning_effort: None,
+            plan_mode: false,
         };
 
         let response = self.driver.complete(request)