ci: drizzle-mysql-smoke job — run #489 fixture against real MySQL (#804)

[proggeramlug]

What

Closes #804 — wires the tests/release/packages/drizzle-mysql/ fixture (added in #2160) into CI against a real MySQL.

Adds a new drizzle-mysql-smoke job to .github/workflows/test.yml with a services: mysql:8 block. MYSQL_ALLOW_EMPTY_PASSWORD=yes and MYSQL_DATABASE=perry_drizzle_test match the fixture's expectations (root with no password, database pre-created). The job builds perry, waits for MySQL to accept connections, then invokes bash tests/release/packages/drizzle-mysql/fixture.sh.

Why services-block (and not Docker in the fixture)

scripts/release_sweep_tiers/tier03_real_packages.sh documents the project preference: fixtures themselves use mock/embedded backends — no Docker in fixture.sh. That's about the local sweep on macOS. CI is a separate context: GH Actions runs ubuntu-latest, the standard idiom for getting MySQL reachable on the runner is a services: block (Docker behind the scenes, but transparent to the fixture). The fixture remains unchanged — it still skips gracefully when no mysqld is reachable, which is what happens on a macOS dev box without docker.

Gating

Mirrors compile-smoke / doc-tests / parity:

• runs on tag push (github.event_name == 'push'),
• runs on workflow_dispatch with run_extended_tests=true,
• opt-in on a PR via the run-extended-tests label.

Real-DB setup + perry release build + drizzle + @perryts/mysql compile is ~15 min wall — PRs shouldn't pay it by default. Release tags still get the gate before publish.

Verification

• YAML parses (python -c "import yaml; yaml.safe_load(open('.github/workflows/test.yml'))" → 10 jobs, drizzle-mysql-smoke present with 8 steps + 1 service).
• Fixture re-run locally: PASS drizzle-mysql (the local mysqld on 127.0.0.1:3306 path the fixture already handled).
• The CI run for this PR (with the run-extended-tests label) is the actual end-to-end check — it's the first time real MySQL has run in this repo's CI.

#489 status after this lands

Together with #2160, both #489 (local e2e) and #804 (CI version) close. The compiler/runtime path through drizzle + @perryts/mysql + real MySQL is verifiable both locally and in CI.

Workaround note carried forward: the fixture entry uses explicit .execute() to dodge #2159 (inherited .then on drizzle's MySqlSelectBase returns undefined under perry). Drop the .execute() calls when #2159 lands.