ai: mirror chat model on local client to prevent mid-turn model swap

When the user selects a model from the picker, the local client's
primary_model and vision_model now match the chat model. Prevents
llama-swap exclusive mode from swapping models when describe_photo
or rerank fires during an agentic turn.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

src/ai/insight_generator.rs

@@ -3635,8 +3635,16 @@ Return ONLY the summary, nothing else."#,
         };
 
         // ── local client (utility calls: rerank, describe_image, etc.) ─
+        // For llamacpp: mirror the chat model selection so rerank /
+        // describe_image hit the same model that's already loaded —
+        // avoids a mid-turn model swap in llama-swap exclusive mode.
         let local: Box<dyn LlmClient> = if local_via_llamacpp {
-            Box::new(self.llamacpp.as_ref().unwrap().as_ref().clone())
+            let mut lc = self.llamacpp.as_ref().unwrap().as_ref().clone();
+            if let Some(ref m) = overrides.model {
+                lc.primary_model = m.clone();
+                lc.set_vision_model(m.clone());
+            }
+            Box::new(lc)
         } else {
             Box::new(self.ollama.clone())
         };