fix: save display_text instead of full_response to conversation logs

Fixes bug where only user input was being logged in session saves.

The issue was that we were displaying 'display_text' (harmony-formatted
response) to the user, but saving 'full_response' (raw text) to logs.

With harmony responses, if token extraction doesn't work but display
does, full_response could be empty while display_text has content.

Changes:
- Save display_text to conversation_history (line 1485)
- Update self.last_response to use display_text (line 1366)
- Ensures logs match what user sees on screen

This makes the behavior consistent: what you see is what gets saved.

Author: jwesleye

src/basic_agent_chat_loop/chat_loop.py

@@ -1333,7 +1333,6 @@ async def _stream_agent_response(self, query: str) -> dict[str, Any]:
 
             # Render collected response
             full_response = "".join(response_text)
-            self.last_response = full_response
 
             # Process through Harmony if available
             display_text = full_response
@@ -1363,6 +1362,9 @@ async def _stream_agent_response(self, query: str) -> dict[str, Any]:
                 if processed.get("has_tools"):
                     logger.debug("Harmony response contains tool calls")
 
+            # Store last response for copy commands (what user sees)
+            self.last_response = display_text
+
             if self.use_rich and display_text.strip() and self.console:
                 # Use rich markdown rendering
                 print()  # New line after agent name
@@ -1482,7 +1484,7 @@ async def _stream_agent_response(self, query: str) -> dict[str, Any]:
                     {
                         "timestamp": time.time(),
                         "query": query,
-                        "response": full_response,
+                        "response": display_text,  # Save what user sees (includes harmony formatting)
                         "duration": duration,
                         "usage": usage_info,
                     }