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/llamacpp.rs

@@ -64,6 +64,12 @@ pub struct LlamaCppClient {
     top_p: Option<f32>,
     top_k: Option<i32>,
     min_p: Option<f32>,
+    /// When `Some`, forwarded to llama-server as
+    /// `chat_template_kwargs: {"enable_thinking": <bool>}`. The Jinja chat
+    /// template (e.g. Qwen3) reads this to gate its reasoning block. `None`
+    /// omits the key entirely, leaving the template's own default. Templates
+    /// that don't reference the key ignore it, so sending it is harmless.
+    enable_thinking: Option<bool>,
 }
 
 impl LlamaCppClient {
@@ -89,6 +95,7 @@ impl LlamaCppClient {
             top_p: None,
             top_k: None,
             min_p: None,
+            enable_thinking: None,
         }
     }
 
@@ -104,6 +111,12 @@ impl LlamaCppClient {
         self.num_ctx = num_ctx;
     }
 
+    /// Set the reasoning toggle forwarded as `chat_template_kwargs.enable_thinking`.
+    /// `None` leaves the chat template's own default in place.
+    pub fn set_enable_thinking(&mut self, enable_thinking: Option<bool>) {
+        self.enable_thinking = enable_thinking;
+    }
+
     pub fn set_sampling_params(
         &mut self,
         temperature: Option<f32>,
@@ -458,6 +471,12 @@ impl LlamaCppClient {
         // via -c, so we silently drop the override here. The config.yaml
         // entry is the source of truth for context size.
         let _ = self.num_ctx;
+        // Reasoning toggle for thinking-capable templates (Qwen3 et al.).
+        // llama-server forwards chat_template_kwargs into the Jinja render
+        // (requires --jinja); templates that ignore the key are unaffected.
+        if let Some(think) = self.enable_thinking {
+            v.push(("chat_template_kwargs", json!({ "enable_thinking": think })));
+        }
         v
     }