docs: add dataset loader and reward function tutorial

[O7c2]

What does this PR do?

Add a step-by-step cookbook tutorial (writing-loaders-and-rewards.rst) that teaches users how to write custom
dataset loaders and reward functions for AReno. Covers the function signatures, record shapes per training mode
(SFT/DPO/RL/Agentic), worked examples for both prompt-based RL and agentic RL, debugging tips for isolated testing,
and common pitfalls.

Also adds cross-references from the existing concept guides (concepts/dataset-formats.rst,
concepts/reward-functions.rst) to the new tutorial.

Related issue

Fixes #42

Type of change

• 📝 Documentation update

How was it tested?

Built docs with sphinx-build -b html -W . _build/html — zero warnings. Verified all :doc: cross-references resolve
correctly.

Checklist

• The PR title summarizes the contribution.
• Linked the related issue in the description (if any).
• Existing tests pass (pytest tests/ -k cpu). (skipped — doc-only change)
• New behavior is covered by tests. (N/A — doc-only change)
• Described the test commands run and any hardware limitations.
• Public API / CLI changes are additive and backward-compatible (see CONTRIBUTING.md). (N/A)

---

[O7c2]

#42