fix(queue): make Redis delivery crash-recoverable

[abnegate]

What changed

• add an idempotent publisher capability with caller-stable message IDs, typed Enqueued / Existing acknowledgements, and a typed conflict for mismatched reuse
• claim pending Redis messages and register processing state atomically with Lua
• add opt-in visibility leases, explicit renewal, timeout reclaim, and receipt-fenced acknowledgements
• keep Redis Cluster keys in the pending queue's hash slot and preserve ordinary enqueue, priority, failed retry, reconnect polling, and legacy in-flight acknowledgement paths
• replace the single-architecture Redis Cluster test image with an official multi-architecture Redis fixture

Why

The previous Redis receive path removed an envelope from the pending list before storing its job and processing records. A worker or connection failure in that gap could strand the message. Producers also could not safely retry a publish after losing its response because the broker generated a new ID for every enqueue.

enqueueOnce() now lets a producer retry the same canonical envelope safely. Atomic claims remove the pending-to-processing loss window. Visibility reclaim remains disabled by default for compatibility and is enabled only when the caller chooses a processing deadline; long-running workers can renew that lease explicitly.

Validation

• bin/monorepo validate
• XDEBUG_MODE=off bin/monorepo check queue
• XDEBUG_MODE=off bin/monorepo test queue — 35 unit tests / 98 assertions; 33 e2e tests / 169 assertions
• real standalone Redis and three-node Redis Cluster coverage for producer retry, conflict, priority, response loss, acknowledgement loss, claim reclaim, lease renewal, reconnect, retry, concurrency, and cluster hash-slot compatibility
• Vale 3.15.1 on packages/queue/README.md

Linear: DAT-1993

[abnegate]

@greptile review

[greptile-apps]

Greptile Summary

Makes Redis queue delivery crash-recoverable.

• Adds idempotent publication with stable message IDs and conflict detection.
• Moves message claims, acknowledgements, rejection, renewal, and retry operations into atomic Lua scripts.
• Adds visibility leases, receipt fencing, timeout reclaim, and Redis Cluster-compatible key derivation.
• Extends pooled and locking connections with the new publishing, leasing, and Lua capabilities.
• Adds standalone and clustered Redis durability coverage and updates the test fixture.

Confidence Score: 5/5

The PR appears safe to merge because no additional blocking failure eligible for this follow-up review was identified.

No blocking failure remains within the scope of the previous review threads.

Important Files Changed

Filename	Overview
packages/queue/src/Queue/Broker/Redis.php	Implements atomic Redis queue claims, receipt-fenced acknowledgements, visibility leases, idempotent publication, and retry compatibility.
packages/queue/src/Queue/Broker/Redis/Keys.php	Derives queue-specific Redis keys that remain colocated within a Redis Cluster hash slot.
packages/queue/src/Queue/Connection/Redis.php	Adds Lua-script execution support to the standalone Redis connection.
packages/queue/src/Queue/Connection/RedisCluster.php	Adds cluster-aware Lua-script execution support.
packages/queue/src/Queue/Connection/Locking.php	Serializes Lua evaluation when Redis connections are shared between coroutines.
packages/queue/src/Queue/Message.php	Adds delivery receipts to message serialization for acknowledgement fencing.
packages/queue/docker-compose.yml	Replaces the previous Redis Cluster image with a three-node official Redis test fixture.
packages/queue/tests/Queue/E2E/Adapter/RedisDurabilityTest.php	Adds standalone Redis coverage for publication and delivery durability behavior.
packages/queue/tests/Queue/E2E/Adapter/RedisClusterDurabilityTest.php	Adds Redis Cluster durability and hash-slot compatibility coverage.

_{Reviews (4): Last reviewed commit: "fix(queue): make legacy retries idempote..." | Re-trigger Greptile}

[abnegate]

@greptile review

[abnegate]

@greptile review

[abnegate]

@greptile review

[abnegate]

Closing as campaign scope cleanup. Crash-recoverable queue delivery primitives are broad shared infrastructure, not a narrowly evidenced dedicated-database fix. Nothing from this draft was merged or released. A future Queue proposal needs its own independently approved scope and consumers.