Refactor/rl cli

[ryanmccann1024]

📝 Description

This PR refactors the RL training pipeline to restore compatibility with the updated project structure and configuration system. Specifically:

• Updated train_pipeline.py to correctly extract, format, and pass simulation configuration data to the RL environment.
• Patched create_environment() to accept flat config dictionaries and wrap them as needed (e.g. in an s1 key) for SimEnv.
• Preserved backwards compatibility with SimEnv expectations while maintaining minimal surface area for potential breakage.
• Added TODOs in strategic places to revisit hardcoded assumptions in future refactors.

These changes unblock usage of the run_train.py CLI entrypoint and allow proper training execution using --agent_type rl with .ini-based configs.

---

Fixes

Fixes: # (restore compatibility of RL pipeline post-refactor)

---

🧐 Type of Change

• 🐛 Bug fix (non-breaking change which fixes an issue)
• ✨ New feature (non-breaking change which adds functionality)
• 💥 Breaking change (modifies pipeline behavior and SimEnv input expectations)
• 💅 UI/UX change (improves the user interface or experience)
• 📚 Documentation update
• ⚙️ Chore (refactoring, compatibility patching)

---

💥 Impact of Breaking Change

• SimEnv still expects sim_dict to include an s1 key. This PR introduces a shim inside create_environment() to enforce that format.
• Future agents or environments expecting different formats may require additional preprocessing or abstraction.
• Config management and training execution now rely on ConfigManager.from_args() and support .ini configs end-to-end.

---

🧪 How Has This Been Tested?

All changes were tested by:

• Running the full RL training CLI using:

  python -m fusion.cli.run_train --config ini/run_ini/config.ini --agent_type rl --run_id test123

[ryanmccann1024]

Again, until we finalize all CLI entries should we worry about the Cross-platform pipeline?

I think it should be the last step and then we check. But, I could be completely wrong.