Support project cache invalidation across nodes in cluster mode

[minwoo1999]

What would you like to be added:
A mechanism to broadcast and propagate project cache invalidation events across all nodes in a sharded cluster.

Why is this needed:
Currently, each node maintains its own in-memory project cache. While cache invalidation in the interceptor works for single-node deployments, it does not guarantee consistency in cluster mode since other nodes still retain stale cache. To ensure consistent behavior across nodes, especially after operations like API key rotation or project updates, a distributed cache invalidation mechanism or inter-node broadcast system is necessary.

See also: sharded-cluster-mode.md

This issue is derived from #1296.

Additional context:
For now, leaving a TODO or design consideration for this enhancement would be helpful for future implementation planning.

[krapie]

Well, It seems like we’re ending up with a implementing additional broadcasting system in our cluster servers, even though we migrated from a broadcast-based cluster to a shard-based one 😅