Add canonical peer status logs

[MarcoPolo]

This PR logs 1% of the following events:

1. A connection is established.
2. A handshake fails.

This is useful for a couple reasons:

1. High churn from a misbehaving peer becomes obvious. If a peer churns 100 connections / second. Then it'll log roughly once a second. An operator can either manually ban this peer or hook the logs up to fail2ban so that the system automatically blocks misbehaving peers. See Add canonical peer status logging with sampling go-libp2p-core#269 for an example.
2. Handshake failures in aggregate also become obvious.
3. Keep a pulse on the system. e.g. No new connections in hours seem suspicious.
4. At a glance you can also see the distribution of quic vs the rest of the transports. This would be better as a metric of course, but still interesting.

The tests will fail since this updates go-libp2p-core which breaks go-libp2p-circuit, which would work if we updated go-libp2p-circuit's dependency on go-libp2p.

Not sure how to resolve. One idea is to depend on a branch of go-libp2p-circuit which depends on go-libp2p-master. Any other suggestions?

[marten-seemann]

We can escape dependency hell by rebasing after #1626 was merged.

[marten-seemann]

LGTM, modulo the leftover log in QUIC.