Consider prometheus histogram metrics in pool

[alexvanin]

Request time distribution looks kinda cool: nspcc-dev/neo-go#2750
Probably we can adopt metric interface to provide simple histogram data to service apps.

[alexvanin]

Estimate possibilities of these changes:

• We can return actual request durations from pool instead of avg.
• We can have callback for each pool request.
• We can discuss tight integration of prometheus and pool (backup plan).

[KirillovDenis]

return actual request durations

In this case I suppose we can have only last n requests durations for each methods (we can configure n in pool params)

have callback for each pool request

We can define something like this:

type RequestCallback func(RequestInfo)

type RequestInfo struct {
	Address    string
	MethodName string
	Elapsed    time.Duration
}

Both options are possible and require almost the same efforts.
Callback seems more convenient

[alexvanin]

Callback implementation looks very nice. Let's go for that.