Recursive dialing can block

Nobody has claimed this yet.

If two peers advertise each other as, e.g., relays and a third peer tries to dial both of these peers at the same time, the dial will bock.

* Thread 1: Start dial to peer A
* Thread 2: Start dial to peer B
* Thread 1: Try to dial the relay (B) for peer A.
* Thread 2: Try to dial the relay (A) for peer B.

Now, given that we allow 8 parallel dials per peer (default), this shouldn't be the end of the world. In most cases, we'll _also_ try dialing peer A and B directly, if we have the appropriate addresses.

Also note, each peer will timeout dialing the relay after _60_ seconds because that's the default _transport_ dial timeout.

However, we should still provide a solution for this.

~Initially reported in https://github.com/libp2p/go-libp2p-swarm/issues/161.~

Start by tracing the two-peer relay-dial scenario described in the issue, including the 8-parallel-dials-per-peer limit and the 60-second transport timeout. Reproduce simultaneous dials to peers A and B, then identify how recursive relay attempts become mutually blocked. Done should provide a solution that prevents this blocking behavior without relying on timeouts.

Written by the indexing model from the issue text.