fix: Add None check in _trim_messages to prevent TypeError

Prevent TypeError when comparing int with NoneType in message trimming.

Problem:
When agent execution starts, _trim_messages() is called to manage context.
The method calculates max_tokens = context_window - reserve_tokens, then
compares current_tokens <= max_tokens. If context_window or reserve_tokens
is None (edge case with model initialization), this causes:
  TypeError: '<' not supported between instances of 'int' and 'NoneType'

Solution:
Add safety check after getting context_window and reserve_tokens.
If either value is None, log warning and return early without trimming.

This prevents the crash while maintaining backward compatibility - the
trimming simply won't happen in edge cases where context values cannot
be determined, which is safe (agent will continue without context management).

Error location: agentmesh/protocol/agent_stream.py line 387
Error message from user:
  [ERROR][2026-02-03 21:13:07][agent_stream.py:162] - Agent execution error:
  '<' not supported between instances of 'int' and 'NoneType'

Author: cowagent

agentmesh/protocol/agent_stream.py

@@ -373,6 +373,12 @@ def _trim_messages(self):
         # Get context window and reserve tokens from agent
         context_window = self.agent._get_model_context_window()
         reserve_tokens = self.agent._get_context_reserve_tokens()
+        
+        # Safety check: ensure we have valid values
+        if context_window is None or reserve_tokens is None:
+            logger.warning("Cannot trim messages: context_window or reserve_tokens is None")
+            return
+        
         max_tokens = context_window - reserve_tokens
 
         # Estimate current tokens