fix(backend): pass dry_run to add_graph_execution in AgentExecutorBlock

The sub-agent's graph validation rejects missing credentials. During
dry-run, credential errors should be stripped — but the dry_run flag
wasn't being passed to add_graph_execution, so validation always
enforced credentials even in dry-run mode.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

autogpt_platform/backend/backend/blocks/agent.py

@@ -88,6 +88,7 @@ class AgentExecutorBlock(Block):
             execution_context=execution_context.model_copy(
                 update={"parent_execution_id": graph_exec_id},
             ),
+            dry_run=execution_context.dry_run,
         )
 
         logger = execution_utils.LogMetadata(