docs(cookbooks): formalize the RL training guide

[ChinmayK0607]

What changed

• Reorganized the RL training cookbook around setup, runnable commands, training flow, trajectory transport, and loss behavior.
• Replaced conversational and fragment-heavy prose with concise technical explanations.
• Added reference tables for runtime trajectory handling and built-in losses.
• Clarified the custom-loss execution path, supported inputs, and value-head limitation without changing the examples or APIs.

Why

The existing README was dense and difficult to scan. It mixed setup, architecture, and implementation details, which made the cookbook harder to follow for readers using it as a training guide.

Impact

The cookbook now provides a clearer path from setup to a first run, while preserving the details needed to understand remote and local trajectories, built-in objectives, and custom policy-gradient losses.

Validation

• Documentation-only change; no runtime behavior changed.
• Reviewed the rendered README in the local cookbook preview.
• Checked the final diff for whitespace errors and removed conversational filler and em dashes.

Made with Cursor