swarm: simplify how we classify address as FD consuming

We need to classify addresses as FD/Non-FD consuming at two places:

• Before we dial them. because we rate limit FD consuming addresses:

  • Relay addresses should be classified as Non-FD consuming for this purpose as the FD will be consumed later when the Relay code passes us the address of the Relay server.

• When We partition the addresses into FD/Non-FD consuming and pass them to the dial code in two batches.

  • This ensures we attempt to dial the FD consuming addresses ONLY after we've exhausted all the FD consuming addresses.
  • However, for this purpose, we need to classify circuit relay addresses based on the address of the relay server so we can pass it to the appropriate batch.

Note: An FD consuming address for now is a TCP or WebSocket address.

See https://github.com/libp2p/go-libp2p-swarm/pull/212 and https://github.com/libp2p/go-libp2p-swarm/pull/212#discussion_r421816941 for the motivation.

No files or tests are named. Start by tracing the two address-classification sites described in the issue, along with the circuit relay dialing path and the motivation in PR #212. Done means relay addresses are treated as non-FD-consuming before dialing, while batching classifies them using the relay server address; TCP and WebSocket addresses remain FD-consuming.