Use a dedicated threadpool for tracing rpc requests

[mattsse]

Describe the feature

all rpc blocking tasks is currently executed on the same threadpool which is fifo, once saturated with a lot of long running tracing requests, this can cause latencies for other, cheaper rpc requests that require blocking io, like eth_ namespace.

tracing is bottlenecked by CPU and perhaps a dedicated rayon threadpool is more suitable here.

ref #3896

Additional context

No response