Add support for CLUSTER SHARDS command

[mp911de]

Allows us to replace our CLUSTER NODES approach for topology discovery.

See redis/redis#10293

[mp911de]

For the time being, we will stick to CLUSTER NODES to avoid incompatibilities in requiring a newer Redis version. It would make sense to have a config flag in ClusterClientOptions that determines whether to use CLUSTER NODES vs CLUSTER SHARDS. The latter command comes with replication-offset details allowing to skip nodes that are not yet fully replicated.

[mp911de]

Note: PooledClusterConnectionProvider uses Partitions.getPartitionBySlot(…) for getWriteConnection which may fail because getPartitionBySlot does not distinguish between replicas and master nodes.

[srgsanky]

If Lettuce supports CLUSTER SHARDS, we can make use of its health field to filter out replicas that are LOADING. The health is determined based on the replication offset which is exchanged by nodes as part of gossip. But #2488 mentions that CLUSTER SHARDS does not expose MYSELF.