Add combined monitor

[Evian-Zhang]

Related to issue #2895. Now we can create multiple monitors for one fuzzer instance.

Currently, the API of Monitor is not good enough to implement a zero-cost combined monitor, the client stats need to be synced to underline monitors in the right time (I choose to sync before each display call), which incurs some runtime overhead.

[tokatoka]

thank you

[domenukk]

Yeah, we probably don't want to clone every time. We should probably "just" forward everything to each sub-monitor(?)

[Evian-Zhang]

@domenukk Yes, but since the API of monitor could give user the mutable reference of client stat, we have no idea of when user changes the data, unless we make the underline client_stats of both sub-monitors point to the same vector, which is impossible using current Monitor API.

[Evian-Zhang]

A possible solution may be returning a custom structure instead of bare mutable reference, which implements Drop trait, and notify the Monitor when it is dropped. But this is a breaking change, and if you think this is acceptable, I'm willing to write a PR when I have time:)

[domenukk]

How about an abstraction that is more event-based?
Then we could even forward everything to monitors via socket or similar at some point

[Evian-Zhang]

@domenukk In my understanding, a simple event-based API would look like

trait Monitor {
    fn on_stat_changed(&mut self, client_id: ClientId, stat: ClientStats);
}

Although this API is good for CombinedMonitor to work, however, since ClientStats is a relatively large structure, forwarding it around may be not a good idea for other normal monitors.

If we want to make the whole thing pure, just make it like a monad:

trait Monitor {
    fn on_stat_changed<F: Fn(&mut ClientStats)>(&mut self, client_id: ClientId, update: F);
}

This approach may be more efficient, and we can still forward everything to Monitors via socket using this API.

What's your advice?