fix(desktop): resolve 2 release-blocking P1 defects

P1-04: GenerationPipeline hardcoded model="default" causing classroom
generation 404. Added model field to GenerationPipeline struct, passed
from kernel config via with_driver(driver, model). Static scene
generation now receives model parameter.

P1-03: LLM API concurrent 500 DATABASE_ERROR. Added transient DB error
retry (PoolTimedOut/Io) in create_relay_task with 200ms backoff.
Recommend setting ZCLAW_DB_MIN_CONNECTIONS=10 for burst resilience.

crates/zclaw-kernel/src/generation/mod.rs

@@ -248,6 +248,7 @@ pub struct GenerationPipeline {
     scenes: Arc<RwLock<Vec<GeneratedScene>>>,
     agents_store: Arc<RwLock<Vec<AgentProfile>>>,
     driver: Option<Arc<dyn LlmDriver>>,
+    model: String,
 }
 
 impl GenerationPipeline {
@@ -265,12 +266,14 @@ impl GenerationPipeline {
             scenes: Arc::new(RwLock::new(Vec::new())),
             agents_store: Arc::new(RwLock::new(Vec::new())),
             driver: None,
+            model: "default".to_string(),
         }
     }
 
-    pub fn with_driver(driver: Arc<dyn LlmDriver>) -> Self {
+    pub fn with_driver(driver: Arc<dyn LlmDriver>, model: String) -> Self {
         Self {
             driver: Some(driver),
+            model,
             ..Self::new()
         }
     }
@@ -353,7 +356,7 @@ impl GenerationPipeline {
                 let item = item.clone();
                 async move {
                     if let Some(d) = driver {
-                        Self::generate_scene_with_llm_static(d.as_ref(), &item, i).await
+                        Self::generate_scene_with_llm_static(d.as_ref(), &self.model, &item, i).await
                     } else {
                         Self::generate_scene_for_item_static(&item, i)
                     }
@@ -413,7 +416,7 @@ impl GenerationPipeline {
         request: &GenerationRequest,
     ) -> Result<Vec<OutlineItem>> {
         let llm_request = CompletionRequest {
-            model: "default".to_string(),
+            model: self.model.clone(),
             system: Some(self.get_outline_system_prompt()),
             messages: vec![zclaw_types::Message::User {
                 content: prompt.to_string(),
@@ -469,6 +472,7 @@ Use Chinese if the topic is in Chinese. Include vivid metaphors and analogies."#
 
     async fn generate_scene_with_llm_static(
         driver: &dyn LlmDriver,
+        model: &str,
         item: &OutlineItem,
         order: usize,
     ) -> Result<GeneratedScene> {
@@ -488,7 +492,7 @@ Use Chinese if the topic is in Chinese. Include vivid metaphors and analogies."#
         );
 
         let llm_request = CompletionRequest {
-            model: "default".to_string(),
+            model: model.to_string(),
             system: Some(Self::get_scene_system_prompt_static()),
             messages: vec![zclaw_types::Message::User {
                 content: prompt,