record latency for pipeline & transaction

[soloestoy]

In redis' protocol, we allow developers using pipeline, but if developers put too many commands into one pipeline there will be latency.

And in replication master pipeline to slave, because we cached the write commands until eventLoop ends.

Also happens in multi/exec.

So, I think it's necessary to record latency for pipeline and transaction, that can help developers to find if the pipeline and transaction are misused.