feat(learning): add the workers and redis job queue roadmap

[OthmaneZ05]

Summary of Changes

Adds the third catalog roadmap, “Workers & the Redis job queue” (redis-queue-workers, EN + FR, 8 steps, ~40 min, intermediate) in the Nimbus Books universe. The learner feels a 2-second synchronous checkout, decouples it with a producer pushing orders onto a real Redis list, writes a blocking BRPOP worker, archives receipts into PostgreSQL (stateless workers / shared state), floods the queue with a 300-order flash sale to watch backpressure live, scales consumption with an auto-scaling group cloned from the worker (a security-group rule with the ASG node as Source covers every replica), and finishes with a poison message handled by retry counters shared in Redis plus a dead-letter queue. Step 7 is an explicit "validate before fixing" beat, and every validator was chosen to remain stable when earlier steps are re-validated after completion.

Uses existing validators only: container_running, http_get_contains, edge_exists, redis_key_exists, table_exists, port_denied, asg_replicas.

Types of Changes

• New feature / node type addition
• Bug fix (non-breaking change resolving an issue)
• Refactoring / structural cleanup
• Documentation update

Verification & Testing

Automated Checks

• Run npm run lint successfully with no errors
• Run npm run build successfully with no compilation errors
• Run npm test successfully (all tests pass) — backend 400/400

Manual Verification

npm run roadmap:validate OK for both language files; a script asserts EN/FR ids and validators are strictly identical.

Played the full roadmap against real Docker with an API driver that mirrors the canvas exactly (network config with default subnet routes, per-node security groups, ASG deploy via the same endpoints as the modal) and executes in the containers the exact code blocks extracted from the roadmap JSON. Result: 34/34 checks green —

• every step validated fail first, then pass (e.g. the producer step fails while the old synchronous code runs; the flash-sale step fails during the flood, before the ASG);
• flash sale: 300 orders queued in seconds, backlog drained to 0 by worker-1 + 3 ASG replicas; SELECT worker, COUNT(*) FROM receipts GROUP BY worker shows 4 distinct container hostnames (67/99/67/68);
• poison message: 3 shared-counter retries then parked in queue:orders:dead, checkout still accepting;
• revisit stability: all 8 steps re-validated green after completion;
• kill & restart mid-journey: web then worker-1 stopped/restarted one at a time (canvas-style) — servers revive via the baked boot line and all 8 steps re-validate green;
• catalog lists both language variants.

Adjacent findings (not addressed here, per scope rules)

1. The derssa/backend-lab-mongo:v1 image is currently unusable on a Fedora 44 / kernel 7.0.9 host: mongod silently segfaults (exit 139) 30–90 s after boot, including a bare docker run outside Torollo (reproduced 3×). This is why the receipt archive uses PostgreSQL (table_exists) instead of the originally considered Mongo (mongo_collection_exists stays unexercised by the catalog). The image likely needs a rebuild on a newer mongod.
2. Stopping two containers simultaneously and restarting them can reshuffle their subnet IPs; peer containers' firewall rules then keep the stale source IPs even after a config re-save (the persisted nodeIpMap wins over live state), silently breaking cross-node traffic. A single-container stop/start (the canvas gesture) keeps its IP and re-wires correctly — verified. Worth a backlog entry.

Checklist

• My code follows the repository's code style and lint standards
• I have updated the documentation or instructions if necessary
• All unit and integration tests are passing