Capture reasoning/thinking from API stream into Thinking entries

StreamResult now includes accumulated reasoning text. After each
stream completes, if reasoning was produced, a Thinking entry is
pushed to the conversation before the response message.

Reasoning content is visible in the context tree UI but not sent
back to the API and doesn't count against the token budget.

Co-Authored-By: Proof of Concept <poc@bcachefs.org>

src/agent/api/mod.rs

@@ -578,6 +578,8 @@ pub(crate) struct StreamResult {
     pub(crate) error: Option<String>,
     /// Remaining display buffer (caller should flush if not in a tool call).
     pub(crate) display_buf: String,
+    /// Accumulated reasoning/thinking content.
+    pub(crate) reasoning: String,
     /// Whether we were mid-tool-call when the stream ended.
     pub(crate) in_tool_call: bool,
 }
@@ -601,6 +603,7 @@ pub(crate) async fn collect_stream(
     let mut error = None;
     let mut first_content = true;
     let mut display_buf = String::new();
+    let mut reasoning_buf = String::new();
     let mut _streaming_guard: Option<super::ActivityGuard> = None;
 
     while let Some(event) = rx.recv().await {
@@ -668,8 +671,7 @@ pub(crate) async fn collect_stream(
                 }
             }
             StreamEvent::Reasoning(text) => {
-                // TODO: reasoning tokens → context entries
+                reasoning_buf.push_str(&text);
-                let _ = text;
             }
             StreamEvent::ToolCallDelta { index, id, call_type, name, arguments } => {
                 while tool_calls.len() <= index {
@@ -696,5 +698,5 @@ pub(crate) async fn collect_stream(
         }
     }
 
-    StreamResult { content, tool_calls, usage, finish_reason, error, display_buf, in_tool_call }
+    StreamResult { content, tool_calls, usage, finish_reason, error, display_buf, in_tool_call, reasoning: reasoning_buf }
 }

src/agent/mod.rs

@@ -477,10 +477,16 @@ impl Agent {
             ).await;
             let api::StreamResult {
                 content, tool_calls, usage, finish_reason,
-                error: stream_error, display_buf, in_tool_call,
+                error: stream_error, display_buf, in_tool_call, reasoning,
             } = sr;
             // --- Stream complete ---
 
+            // Push thinking entry if model produced reasoning
+            if !reasoning.is_empty() {
+                let mut me = agent.lock().await;
+                me.push_entry(context::ConversationEntry::Thinking(reasoning));
+            }
+
             // --- Lock 3: process results ---
             let (msg, pending) = {
                 let mut me = agent.lock().await;