[Bug]When tracing a pipeline, the recorded results are inaccurate

[wanghonglu]

Fred version - 10.1.0
Redis version - 7.0.15
Platform - linux|mac|windows
Deployment type - cluster|sentinel|centralized

when run code like this

    let pip = redis_client.pipeline();
    pip.get::<(), _>("tracing_redis_key").await.unwrap();
    pip.del::<(), _>("tracing_redis_key").await.unwrap();
    let (value, _): (i32, ()) = pip.all().await.unwrap();

In this case, only pip.all will actually interact with the network and send the command to the server. However, the current code records tracing by opening a span for each get/set. However, the get/set here is actually just assembling commands in memory, which takes a very short time. It will be inaccurate, and the all operation does not record span.
Looking at it on Jaeger, it looks like this

I think we should add a separate span of pipelinecmds, and the name can be attached to all the sub-commands,such as cmd.name=GET,DEL but when I look at the code, it seems that a lot of code needs to be changed if this is to be done.