[13.x] Add cluster_aware Redis mode for managed single-endpoint providers

[timmylindh]

Several managed Redis providers - AWS ElastiCache Serverless (Valkey), Upstash Redis, Redis Cloud with the DMC proxy - present a Redis Cluster behind a single endpoint while blocking the CLUSTER topology commands, so the client can't configure itself as a real cluster client. The endpoint still enforces slot semantics: multi-key commands return CROSSSLOT unless every key hashes to the same slot.

Laravel breaks out of the box on these providers. Cache::many(), Cache::putMany(), tagged cache flush(), and every queue push / pop / later / migrate script issue multi-key commands with no slot guarantee.

The existing hash-tag handling from #59533 only fires for real cluster clients - but here the client is standalone and only the server is clustered. There's no runtime signal to detect this, so the signal has to come from config.

This PR introduces a cluster_aware flag, opt-in per connection:

// config/database.php
'redis' => [
    'default' => [
        // ...
        'cluster_aware' => true,
    ],
],

When set, queue and limiter keys get hash-tagged, and multi-key cache operations fall back to per-key paths.

Changes

Connection infrastructure

• Added isClusterAware() - true for real cluster clients or standalone connections with cluster_aware: true.
• Added supportsMultiSlotMultiKeyCommands() - false when the client can't route a multi-key command across slots. Overridden to true on PhpRedisClusterConnection, which routes MGET/DEL/etc. per slot natively.
• $config moved to the base Connection so PredisConnection can read it; Predis constructor and connector updated accordingly (BC-safe default).

RedisQueue & ConcurrencyLimiter

• Keys are hash-tagged whenever isClusterAware() is true, not just for real cluster clients.

RedisStore

• many() falls back to per-key get when ! supportsMultiSlotMultiKeyCommands(); phpredis cluster keeps the native MGET path.
• putMany() falls back to per-key setex when isClusterAware() - MULTI/EXEC can't span slots on any cluster-like endpoint.

RedisTaggedCache

• flush() uses the per-slot fallback when isClusterAware().
• flushValues() uses pipelined per-key DEL when ! supportsMultiSlotMultiKeyCommands(); phpredis cluster keeps the variadic DEL fast path.

Backward Compatibility

• cluster_aware defaults to false. Zero behaviour change for any existing config.
• isCluster() semantics are unchanged - still true only for real cluster clients.
• Real cluster connections return true from isClusterAware() via the isCluster() branch, so every rewired call site takes the same path it did before for them.
• PredisConnection::__construct($client) still works; the new $config parameter is optional.

[github-actions]

Thanks for submitting a PR!

Note that draft PRs are not reviewed. If you would like a review, please mark your pull request as ready for review in the GitHub user interface.

Pull requests that are abandoned in draft may be closed due to inactivity.