Run the suites in the shared ccc test image (environment.kind: shared) instead of the per-project aidev-tests image

Switch this project to aidev's shared test image route (environment.kind: shared). Operator guide: aidev docs/operations/shared-test-images.md. The operator approved widening it on 2026-10-30, after velocity-js-sdk proved the route.

Today .aidev/project.yaml runs every suite in registry.gitlab.syncad.com/hive/velocity-identity/aidev-tests@sha256:71f91fa96a80db336ede94df4510acf4c3b7ca0da0218f7e7dabd9a7a1f9ef32, built by .aidev/runtime/Dockerfile.

This project's integration suite starts PostgreSQL, so it uses ccc's node-test-tools-pg variant. That is node-test-tools plus the PostgreSQL 18 server, published under the same develop tag. Checked 2026-10-30: Node v24.22.1, pnpm 10.33.0; initdb, pg_ctl and postgres are on PATH (/usr/lib/postgresql/18/bin); a postgres user and runuser exist. There is no /opt/pgbin symlink: the old Dockerfile created that. .aidev/run-checks.sh calls /opt/pgbin/pg_ctl and initdb. Make it resolve the PostgreSQL bin directory from PATH (command -v initdb), falling back to /opt/pgbin so the fallback image keeps working. Keep pnpm-workspace.yaml (the catalog symlink into the npm-common-config submodule) in the profile's inputs if aidev's shared-kind schema allows inputs; otherwise say how lockstep still covers it.

Change

Replace the environment section with:

environment:
  kind: shared
  image: registry.gitlab.syncad.com/hive/common-ci-configuration/node-test-tools-pg:develop
  packages:
    lock: pnpm-lock.yaml
    store: /opt/pnpm-store
    fetch: [pnpm, fetch, --frozen-lockfile, --store-dir, /opt/pnpm-store]
    install: [pnpm, install, --offline, --frozen-lockfile, --store-dir, /opt/pnpm-store]   # plus the flags .aidev/pnpm-deps.sh uses
  fallback: registry.gitlab.syncad.com/hive/velocity-identity/aidev-tests@sha256:71f91fa96a80db336ede94df4510acf4c3b7ca0da0218f7e7dabd9a7a1f9ef32

Keep store at /opt/pnpm-store: the fallback image has its packages at that path, and aidev's guide requires the fallback's store to be at store. fetch must be the online pnpm fetch --frozen-lockfile --store-dir /opt/pnpm-store. install must be frozen and offline with --store-dir /opt/pnpm-store, and otherwise the same flags .aidev/pnpm-deps.sh already uses.

The shared image does not set the environment the old Dockerfile set. Make .aidev/pnpm-deps.sh (sourced by .aidev/run-checks.sh) export what the suites need: npm_config_store_dir=/opt/pnpm-store, the other ENV lines of the old Dockerfile (npm_config_manage_package_manager_versions=false, HUSKY=0, DO_NOT_TRACK=1), and HOME=/tmp when HOME is unset or not writable. Suites run with the caller's uid and --network none.

Keep .aidev/runtime/ for now, since the fallback digest names the image it builds; a later change drops both once the shared route has run on the hosts that serve this project. Update .aidev/README.md to describe the shared route and the fallback, including that the fallback must be rebuilt, or dropped, in the same commit as any pnpm-lock.yaml change.

This is the same migration velocity-js-sdk made in velocity-js-sdk#6 (closed) (.aidev/project.yaml, .aidev/pnpm-deps.sh, .aidev/README.md on its aidev/integration branch); use it as the reference. velocity-js-sdk#7 (closed) confirmed the route in a real workflow: the worker logged runs its suites in …node-test-tools@sha256:… (shared …:develop pinned to …) with the host package store.

Acceptance

  • Every verification slot passes on the shared route, with the same test counts as on the current image. Before landing, run each slot by hand inside the shared image with --network none, the store read-only, and the caller's uid, as velocity-js-sdk#6 (closed) did.
  • The switching commit is judged under the approved container image, since environment is protected. Note in the summary that the first shared run happens on the next workflow.