feat: Improve subordinate delegation with loop prevention and depth limits

[TerminallyLazy]

Summary

Improvement to the agent delegation system to prevent infinite loops and over-delegation:

• Decision Framework: Added 3-question evaluation system (Subtask Boundary Check, Context Efficiency Check, Specialization Check) with 2-of-3 rule for delegation decisions
• Depth Limiting: Configurable max_agent_depth parameter (default: 5) with hard validation in call_subordinate
• Loop Prevention: Same-profile delegation warning system to detect and log potential loop scenarios
• Enhanced Awareness: Agents display their delegation depth, remaining levels, and receive warnings when near limits or as subordinates

Changes

Prompts

• prompts/agent.system.tool.call_sub.md: Complete rewrite with decision framework, anti-patterns, and profile selection guidance
• prompts/agent.extras.agent_info.md: Added depth tracking placeholders

Code

• agent.py: Added max_agent_depth: int = 5 to AgentConfig
• python/tools/call_subordinate.py: Added depth limit validation and same-profile warning
• python/extensions/message_loop_prompts_after/_70_include_agent_info.py: Added depth calculation and dynamic warnings

[gjadams]

I like it, Nice, looks to address a long-time existing issue with more than just prompt engineering. might be nice to have the settings.json parameter versus hard coded, but that is just a thought. Glad someone engineered a nicely constructed answer to this using agent config and numerical controls. thank you