[TRAIN-8] No end-to-end tutorial for writing and registering custom reward functions

## Problem

The environments guide explains how to use the built-in math and code environments, including multi-reward support. However, it does not provide an end-to-end example for writing a custom reward function and registering it via a custom environment.

## Suggested fix

Add a "Custom Reward Functions" tutorial covering: implementing a reward function, registering it via a custom environment class, and wiring it into a training config with a working example.