Describe the bug

Given that on redis-benchmark we only replace the tag placeholder ":{tag}" when the --cluster flag is passed, on setups that rely on a proxy to abstract the cluster topology to the client this makes them to be limited by standalone performance ( given that we're always targeting the same hash slot )

To reproduce

Run for example a benchmark with a keyspace length larger that 1 and notice that we will always target the shard with the slot of {tag}. Example of Redis Enterprise with single endpoint setup and running the following benchmark:

redis-benchmark -r 16000 -h <HOST>  -p 12000  -t GET -n 100000000 -d 1024 -P 50 -c 50 

Expected behavior

We should only include the hash tag when the --cluster flag is passed to prevent an easy fix this.