AI: add enable_thinking reasoning toggle plumbed to llama.cpp

New optional SamplingOverride forwarded to llama-server as
chat_template_kwargs.enable_thinking (gates Qwen3-style reasoning
blocks). None leaves the template default; other backends ignore it.
Wired through the agentic-insight and chat-turn request bodies/handlers.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

src/ai/backend.rs

@@ -41,6 +41,10 @@ pub struct SamplingOverrides {
     pub top_p: Option<f32>,
     pub top_k: Option<i32>,
     pub min_p: Option<f32>,
+    /// Reasoning toggle. Only the llama.cpp backend honors it (forwarded as
+    /// `chat_template_kwargs.enable_thinking`); other backends ignore it.
+    /// `None` leaves the model/template default in place.
+    pub enable_thinking: Option<bool>,
 }
 
 impl SamplingOverrides {
@@ -124,6 +128,7 @@ mod tests {
             top_p: None,
             top_k: None,
             min_p: None,
+            enable_thinking: None,
         };
         assert!(!empty.has_sampling());
 
@@ -134,6 +139,7 @@ mod tests {
             top_p: None,
             top_k: None,
             min_p: None,
+            enable_thinking: None,
         };
         assert!(with_temp.has_sampling());
     }