fix: handle ChatMessage objects in log_messages by converting at call sites

- Modified memory.py to convert ChatMessage objects to dicts before calling log_messages
- Kept AgentLogger.log_messages interface clean - only accepts list[dict] as documented
- Fixes TypeError when calling agent.replay(detailed=True)
- Conversion happens at the 2 call sites in ActionStep and PlanningStep replay
- Maintains type safety and clear separation of concerns

This cleaner approach:
- Makes the log_messages interface predictable and type-safe
- Explicitly converts ChatMessage objects where they're used
- Avoids mixing domain objects with logging utilities
- Follows 'be conservative in what you send' principle

Resolves issue where replay(detailed=True) would crash with:
TypeError: 'ChatMessage' object is not iterable

Author: ketan0

src/smolagents/memory.py

@@ -261,13 +261,23 @@ def replay(self, logger: AgentLogger, detailed: bool = False):
             elif isinstance(step, ActionStep):
                 logger.log_rule(f"Step {step.step_number}", level=LogLevel.ERROR)
                 if detailed and step.model_input_messages is not None:
-                    logger.log_messages(step.model_input_messages, level=LogLevel.ERROR)
+                    # Convert ChatMessage objects to dicts for logging
+                    messages_as_dicts = [
+                        msg.dict() if hasattr(msg, 'dict') else dict(msg) 
+                        for msg in step.model_input_messages
+                    ]
+                    logger.log_messages(messages_as_dicts, level=LogLevel.ERROR)
                 if step.model_output is not None:
                     logger.log_markdown(title="Agent output:", content=step.model_output, level=LogLevel.ERROR)
             elif isinstance(step, PlanningStep):
                 logger.log_rule("Planning step", level=LogLevel.ERROR)
                 if detailed and step.model_input_messages is not None:
-                    logger.log_messages(step.model_input_messages, level=LogLevel.ERROR)
+                    # Convert ChatMessage objects to dicts for logging
+                    messages_as_dicts = [
+                        msg.dict() if hasattr(msg, 'dict') else dict(msg) 
+                        for msg in step.model_input_messages
+                    ]
+                    logger.log_messages(messages_as_dicts, level=LogLevel.ERROR)
                 logger.log_markdown(title="Agent output:", content=step.plan, level=LogLevel.ERROR)
 
     def return_full_code(self) -> str: