GeezerVison

Makes the smoke test start faster and check more. It also changes one health check setting customers run, on three services: searcher-0, precise-code-intel-worker, and syntactic-code-intel-worker get a 300 second start period.

• Faster start. Pulls images in parallel (docker compose pull), with up to three attempts, because Docker Hub times out now and then.
• Waits for health, not for a clock. Instead of sleep 90, runs docker compose up --wait --wait-timeout 300. That returns once every container is running, and healthy if it has a health check. It fails if a container reports unhealthy.
• Counts services from the Compose file instead of a hard-coded 25. migrator is left out, because it exits when it is done.
• Fails if any container restarted. This replaces a loop which said "checking every 10s for 5 minutes" but checked twice, one second apart.
• Saves every container's logs on failure. Before, it saved logs only for containers listed by docker ps, which hides the ones that exited.
• Uses docker compose (v2) instead of docker-compose (v1). The VM image has v2.18.1.

What customers will see

searcher-0, precise-code-intel-worker, and syntactic-code-intel-worker get start_period: 300s on their health checks, the same as sourcegraph-frontend-internal. searcher-0 had no start period, and the two workers had 60 seconds.

None of the three is ready until sourcegraph-frontend-internal is, which waits for the migrator. Until then the health check fails, and three failed checks, 5 seconds apart, mark the container unhealthy. So on every start searcher-0 showed unhealthy, then healthy, and the two workers did the same whenever the start took longer than their 60 seconds. Each now shows starting until its first check passes.

Nothing else changes for customers: Docker does not restart an unhealthy container, and no service waits on the health of these three. docker compose up --wait now works for them too.

Risk: low. One health check setting on three services for customers; the rest is the test.

Things to know

• --wait gives up when any container reports unhealthy, even for a moment. That is why the three services need the start period.
• In build 16853, --wait ran without the start period, and failed with container searcher-0 is unhealthy. Seconds later docker ps showed searcher-0 as healthy.
• On the 8.2.0 images, build 16905 of this PR and build 16902 of ci: Check every Compose file and shell script, and test Jaeger and the executor #1489 failed with container precise-code-intel-worker is unhealthy, 71 seconds after the container started. In both, sourcegraph-frontend-internal started about 30 seconds after the worker, once the migrator was done. The second commit gives the two workers the start period.

Checklist

• All images have a valid tag and SHA256 sum (no image changes)

Test plan

• The smoke test passes in this PR's build: build 16891: passed, the smoke test job took 403 seconds. On the 8.2.0 images, with the start period on the two workers, build 16907 passed in 430 seconds
• docker compose config shows the start period on searcher-0, and 25 services without migrator (Compose 2.18.1 and 5.7.1)
• shellcheck test/smoke-test.sh is clean
• On a test machine (Docker 29.9.3, Compose 5.7.1), with this PR alone on main: the smoke test passes in 35 seconds, 25 of 25 containers, no restarts. main's smoke test takes 2 minutes there
• The same run with searcher-0's start period removed fails after 30 seconds with container searcher-0 is unhealthy, exits 1, and saves 26 log files

Closes FE-586