`cluster/clusterAllReplicas` queries will reuse connections from connection pool

**Use case**

`cluster/clusterAllReplicas` usage with `view(...)`

`view(...)` is quite powerful feature, which allow you to better control which part of query is executed on which node (initiator/shard), but problem is that that cluster wrapper doesn't reuse connections, which is opposite from `Distributed` table, so it makes it less effective in case we have high QPS load

**Describe the solution you'd like**

`cluster/clusterAllReplicas` will reuse connections from connection pool.