Generate a lab from a notebook section

[akash-coded]

The problem

12 labs, ten notebooks, and roughly forty distinct mechanisms taught across them. The
pathway has obvious gaps — nothing on ANN parameters,
nothing on prompt assembly, nothing on agent stopping rules beyond L11's neighbourhood.

Writing a lab is five files and, honestly, most of a day.

The proposal

A scaffolder: python scripts/new_lab.py --track T3 --title "..." writes the five files with the
brief's four beats stubbed, a starter with a NotImplementedError, and a checks file with one
public and one hidden check already wired up.

What it must NOT do

Generate the content. The value of a lab is the decision in the Attribute section, and a
generated decision whose alternatives are obviously bad is a quiz, not a decision. The scaffolder
should refuse to leave the Attribute section empty — a lab with no real choice in it should fail
test_labs.py, and that check is worth writing first.

The harder question underneath

Which mechanisms deserve a lab at all? Not everything does. A lab is worth it when there is a
decision with defensible wrong answers. Where there is one right way, a docs page is better and
cheaper. That triage is the actual work here.