perf/basic_host: Don't handle address change if we hasn't anyone

[snyh]

basic_host.background is used for handling address changed events, but the node may be created with libp2p.NoListenerAddrs.

the h.updateLocalIpAddr is cost a little CPU time.

But I use NoListenerAddrs with many (one thousand) tiny test nodes, and all these test nodes communicate with other peers by a full feature node (by pubsub or relay) in the same process.

The basic_host.background would use many CPU time and add 1000 useless goroutine in this situatuion.

---

I'm not sure whether the network could do listen after host.background started.
So I just keep the background goroutine and skip one round if there hasn't any listener address.

[snyh]

@Stebalien thanks for #398

[Stebalien]

Is it the call to Addrs or the call to update the IP address that costs CPU? Could you post a CPU profile?

Notes:

1. We could easily make Addrs short-circuit if we're not actually listening on any transports. That should save a fair amount.
2. We could avoid updating the local IP address... I'm just slightly concerned about stale addresses and want to see if we can just optimize it.

[snyh]

1.pb.gz

profile with 30s before this PR

[snyh]

2.pb.gz

after this PR

[Stebalien]

After a lot of head scratching to make sure this can't go wrong... I'm convinced it can't go wrong.

[Stebalien]

Well, one small change. We want Addrs and the emitted events to be the same.

[Stebalien]

And thanks for the detailed perf analysis!