fix: gemini route, usage capture, search timeout, and trace finalization

- Add missing /v1/gemini POST route and handler
- Capture MitmEvent::Usage in gemini sync/streaming handlers
- Add retry counter (max 3) to search handler to prevent hang
- Add trace finalization at all gemini_sync channel exit points
- Fix UpstreamError trace outcome label
- Add timeout trace with error recording
- Dispatch Usage before ResponseComplete in SSE flush

src/mitm/intercept.rs

@@ -89,6 +89,8 @@ pub struct StreamingAccumulator {
     pub grounding_metadata: Option<serde_json::Value>,
     /// Buffer for reassembling lines split across TCP reads.
     pub pending_data: String,
+    /// Thinking signature (base64 opaque blob) from non-function-call response parts.
+    pub thinking_signature: Option<String>,
 }
 
 impl StreamingAccumulator {
@@ -150,8 +152,12 @@ impl StreamingAccumulator {
                                         .as_secs(),
                                 });
                             }
-                            // Capture non-thinking response text (skip thoughtSignature parts)
-                            else if part.get("thoughtSignature").is_none() {
+                            // Capture non-thinking response text
+                            else {
+                                // Capture thoughtSignature from response parts (not function call parts)
+                                if let Some(sig) = part.get("thoughtSignature").and_then(|v| v.as_str()) {
+                                    self.thinking_signature = Some(sig.to_string());
+                                }
                                 if let Some(text) = part["text"].as_str() {
                                     if !text.is_empty() {
                                         self.response_text.push_str(text);
@@ -277,6 +283,7 @@ impl StreamingAccumulator {
                 .duration_since(std::time::UNIX_EPOCH)
                 .unwrap_or_default()
                 .as_secs(),
+            thinking_signature: self.thinking_signature,
         }
     }
 }
@@ -302,6 +309,7 @@ fn extract_usage_from_message(msg: &Value) -> Option<ApiUsage> {
             .duration_since(std::time::UNIX_EPOCH)
             .unwrap_or_default()
             .as_secs(),
+        thinking_signature: None,
     })
 }