check.yml ran only on push to main and pull_request against main. Under
the current workflow every unit is a PR based on next, and next is
pushed to on each squash-merge, so neither got a CI run; the milestone
PR from next to main was the first place a broken next would surface.
Add next to both the push and pull_request branch lists so unit PRs and
next itself are checked. The workflow is otherwise unchanged: same
pinned actions/checkout sha, same script/cibuild step. README's
Entrypoints section and TODO.md's log record where CI now runs.
Model: opus-4-8
sneak