docs: correct backpropagation gradient explanation

[Andrew Barnes (Bortlesboat)]

Summary

The lesson now explains backpropagation with the chain rule, so learners no longer see a loss change used as the input to a reversed finite-change equation. It also separates the forward pass, backward gradient computation, and optimizer update.

Fixes #730.

Validation

• Parsed and schema-validated the notebook with nbformat 5.10.4.
• Compared the commit with current main: only markdown cell 29 changed; all code cells, outputs, execution counts, and notebook metadata are unchanged.
• Ran git diff --check successfully.
• Notebook execution was not run because no executable content, dependency, output, or execution order changed.

Post-Deploy Monitoring & Validation

No additional operational monitoring required. This is a markdown-only lesson correction with no runtime path; the PR author and reviewer can validate the equations in GitHub's rendered notebook before merge.

---