ci: run just check on pull requests

[CrazyPandaP]

Stacked on #410 — that PR's three test commits are included here, so this diff will shrink to the workflow file once it lands. I'll rebase it onto main as soon as #410 merges.

The 30 scripts under tests/ are only run by hand today: nothing in .github/workflows/ references tests/, just check or just lint. A change that breaks a test looks green until someone happens to run the suite locally — which is how the three drifts in #410 got in.

This adds a job that runs just check (the 30 scripts plus the Justfile formatting check) on every pull request and on pushes to main.

Facts that shaped it:

• The suite needs bash, python3 and coreutils only — no device, no network, no container. node and podman appear in the scripts only inside strings and stubs.
• It finishes in ~9 seconds on an ordinary x86_64 Ubuntu 24.04 box (measured: 8.76s).
• just comes from extractions/setup-just rather than apt-get, because the Justfile's [group(...)] attributes need just >= 1.22 while Ubuntu 24.04's archive still ships 1.21 — apt-get install just fails with Unknown attribute 'group'.
• just lint is deliberately not wired up here. Its recipe is find . -iname "*.sh" -exec shellcheck "{}" \;, and find does not propagate the exit status of -exec, so the recipe returns 0 no matter what shellcheck reports (it currently reports ~210 findings). Fixing that needs a decision on severity scope or a cleanup pass, so it belongs in its own change.
• The runner is ubuntu-latest (x86_64) because that is where I verified the suite; the tests are arch-agnostic, so switching to ubuntu-24.04-arm for consistency with the other workflows is fine if you prefer.

Happy to adjust anything here — and if the tests are deliberately kept out of CI for a reason I have missed, I would rather know than guess.

[JPyke3]

Love the implementation, happy to add it. I'm just always cautious about over commenting code.

If we could have a quick fix up there I'd appreciate it

[CrazyPandaP]

Applied — the comment is two lines now:

# Controller targets come from per-device ARMADA_IP_TARGETS and are
# filtered against the types supported by controller-type.

I put the change on the tests branch so both PRs carry the trimmed version, then rebased this one on top. just check passes on the result: 30/30 scripts plus the Justfile format check.

[CrazyPandaP]

Applied, and the check running once turned out to be useful — it surfaced two things, both now fixed:

rg: command not found. armada-control-rgb-test.sh uses ! rg -q ARMADA_RGB_ … to assert those variables are gone from the device configs. On a runner without ripgrep, rg exits 127, ! turns that into success, and the assertion passes vacuously. The job now installs ripgrep, and I checked that both assertions really pass once it is there.

armada-store-test.sh — a fourth case of the same drift. The job failed on assert names == {"ARMSX2", "CEMU", "PICO-8_64", "VITA3K", "XENIAEDGE"}: 7490047 ("feat(store): Add BigPEmu, MAME, PrimeHack and Ymir") added the BigPEmu ES-DE find rule but not the assertion that pins that set. That fix is on the tests PR (#410), which is now rebased onto current main.

Both branches are rebased onto current main (behind_by: 0). just check passes locally with ripgrep installed: 30/30 scripts plus the Justfile format check. Would you mind approving the workflow run again when you get a chance?