Update to libp2p-0.36

[romanb]

This PR updates libp2p to 0.36. The full changelog compared to 0.35.1 can be seen here. The highlights are:

• "Parallel negotiation" is re-enabled for multistream-select ([multistream-select] Re-enable 'parallel negotiation'. libp2p/rust-libp2p#1934) and takes effect when the dialer has 3 or more alternative protocols for a particular upgrade. I don't think this case actually ever happens in substrate though, so there should be no surprises.

• libp2p-mdns has undergone a larger refactoring and sends queries much less aggressively (Don't poll network unnecessarily. libp2p/rust-libp2p#1977), which should be compensated for by queries sent whenever a multicast group is joined on a network interface. If there should be unforeseen issues, the query interval can now be configured and could be lowered to the previous 20 seconds.

• There is now a new implementation of the relay protocol, implemented in protocols/relay: Implement circuit relay specification libp2p/rust-libp2p#1838.

• libp2p-dns now uses the trust-dns library ([libp2p-dns] Use trust-dns-resolver (with either async-std or tokio). Remove thread pool. libp2p/rust-libp2p#1927), thereby supporting new features and configuration options as well as removing the previous background thread for performing DNS resolution via std::net::ToSocketAddrs. This also made it possible to implement support for /dnsaddr addresses ([libp2p-dns] Implement /dnsaddr resolution. libp2p/rust-libp2p#1931).

Given the larger changes to the (M)DNS libraries, a burn-in may be appropriate.

polkadot companion: paritytech/polkadot#2672

[tomaka]

Note: I had to run cargo update -p data-encoding -p futures:0.3.12 -p winapi:0.2.8 -p futures-lite to get Polkadot to build.

[tomaka]

Also cargo update -p fs-swap -p winapi:0.2.8, otherwise you get compilation errors when building fs-swap.

[tomaka]

Burn-in looks fine.

One thing that is curious is that before this PR we had little peaks to the number of libp2p tasks, presumably because of the periodic Kademlia requests. Now these peaks are gone.

The same shows on the number of connections. There are still little peaks of number of connections every 10 minutes, when Kademlia requests trigger, but they are way less pronounced.

Anyway, everything seems to work as intended. I think we need a Polkadot companion PR however.

[romanb]

I think we need a Polkadot companion PR however.

Can we repurpose paritytech/polkadot#2659 to serve as the companion?

[tomaka]

I have no idea who to ask for reviews 🙃

[tomaka]

The Polkadot companion needs two approvals as well.

EDIT: as long as I don't restart the CI check, probably no.

[tomaka]

bot merge

[ghost]

Trying merge.