system_cleanup queue-metadata cleanup is not unit-testable without Redis (add Redis to CI or refactor to be generic)

Summary

Follow-up to #816 (fix for #815). The system_cleanup handler that prunes Redis
queue metadata currently has ~0% patch coverage on its core branch, and there is
no straightforward way to unit-test it today. This issue tracks making that code
path testable.

Background

#816 changed handle_cleanup_request in
src/jobs/handlers/system_cleanup_handler.rs to source the Redis pool from the
queue backend instead of the transaction repository, so cleanup runs whenever
the queue backend is Redis regardless of REPOSITORY_STORAGE_TYPE. Codecov
flagged the changed lines as uncovered.

Why it's hard to cover right now

All of the changed lines sit behind the backend_type() == Redis guard, and the
only way to obtain a non-None redis_connections() is through Queue::setup(),
which opens live Redis ConnectionManagers. Consequently:

• A plain unit test cannot reach those lines without a running Redis.
• CI has no Redis service in the coverage job (.github/workflows/ci.yaml), and
  the existing *_redis.rs tests are all #[ignore]d, so a Redis-gated
  integration test would not run in CI (and thus wouldn't contribute coverage).
• handle_cleanup_request takes a concrete ThinData<DefaultAppState> (hard-wired
  to the concrete JobProducer), so a MockJobProducerTrait cannot be injected
  as-is.

Proposed options (any one, or a combination)

1. Make the inner handler generic over the job producer, mirroring the existing
   relayer_health_check_handler pattern (public concrete wrapper +
   generic _impl). This allows injecting MockJobProducerTrait and unit-testing
   the two skip branches (non-Redis backend; Redis backend with no connections)
   in normal CI — covers most of the diff without any infra.

2. Add a Redis service to the coverage CI job and either un-#[ignore] a
   focused integration test or add a new one that exercises the full
   prune path (<queue>:{done,failed,dead} + :data/:data::result).

3. Add a Redis-gated #[ignore]d integration test (mirrors the current
   *_redis.rs convention) that documents/verifies the end-to-end behavior even
   if it doesn't count toward CI coverage.

Start with src/jobs/handlers/system_cleanup_handler.rs and read handle_cleanup_request alongside the relayer_health_check_handler pattern. Then inspect .github/workflows/ci.yaml and the existing *_redis.rs tests to choose between generic unit coverage and Redis-backed integration coverage. Done means the Redis cleanup branches have meaningful test coverage in the intended CI path, with the queue metadata keys exercised where applicable.