fix(ci): retry docker compose up on transient registry pull failures

[AztecBot]

Problem

e2e_ha_full.parallel.test.ts ("should produce blocks with HA coordination and attestations") failed in CI on an unrelated PR (#24492) before the test even started: Docker Hub returned a transient 502 while docker compose up pulled postgres:16-alpine for the HA compose stack.

CI log: http://ci.aztec-labs.com/cac25b9251c99496

postgres Error unknown: failed to resolve reference "docker.io/library/postgres:16-alpine":
unexpected status from HEAD request to https://registry-1.docker.io/v2/library/postgres/manifests/16-alpine: 502 Bad Gateway

This can hit any compose-based test (run_compose_test is shared by the e2e compose/web3signer/ha suites and docs examples) whenever a required image is not in the local Docker cache and the registry blips.

Fix

Wrap the docker compose up -d --force-recreate in ci3/run_compose_test with the existing ci3/retry -p <regex> helper, retrying only failures that match transient registry/network errors (5xx, HEAD-request failures, TLS/connection timeouts, rate limiting). This reuses the same pattern-gated retry convention already used for network flakes in l1-contracts/bootstrap.sh, barretenberg/sol/bootstrap.sh, and noir-projects/aztec-nr/bootstrap.sh, so genuine failures (e.g. an image tag that doesn't exist) still fail immediately without retrying.

Retrying up -d --force-recreate is idempotent: a retry recreates any containers from a partially failed attempt.

Verified the exact error message from the CI log matches the retry pattern (retried, then succeeds), while a manifest not found failure exits immediately without retries.

No tracked issue exists for this flake.

---

Created by claudebox · group: slackbot