Speed up cleanup-removed-hosts-metadata script

[MikhailBurdukov]

Performance improvement for cleanup-removed-hosts-metadata. In this implementation we unite the nodes to delete in transactions to do single operation for batch of nodes.

Script for time comparison:
cat populate.sh

clickhouse client --query "DROP DATABASE IF EXISTS test_db"
clickhouse client --query "CREATE DATABASE test_db"

for i in $(seq 1 $1);
do
    
	
    clickhouse client --query "CREATE TABLE test_db.test$i (a int, b int) engine=ReplicatedMergeTree('/clickhouse/tables/{shard}/test_db.table$i', '{replica}') order by a partition by a"
    clickhouse client --query "SYSTEM STOP MERGES test_db.table$1"
    for j in $(seq 1 $2);
    do
      clickhouse client --query "INSERT INTO test_db.test1 SELECT rand()%10000, rand(1)%1000 FROM numbers(100)"
    done

done
sudo service clickhouse-server stop

Results:
cmd : time chadmin zookeeper cleanup-removed-hosts-metadata <host>

data	new	old
./populate.sh 10 20	0m15.114s	2m46.153s
./populate.sh 20 2	0m4.864s	0m51.093s
./populate.sh 2 40	0m8.673s	5m54.860s
./populate.sh 15 15	2m46.116s	8m42.093s