autonat v2 server tests the last requested address but always reports addr_idx 0

[satyakwok]

Summary

In the AutoNAT v2 server, handle_request_internal selects the address to dial-back with addrs.pop() (the last address in the request) but hardcodes the reported index to 0:

rust-libp2p/protocols/autonat/src/v2/server/handler/dial_request.rs

Lines 275 to 278 in 6a6814b

	all_addrs.clone_from(&addrs);
	let idx = 0;
	let addr = addrs.pop().ok_or(HandleFail::DialRefused)?;
	*tested_addrs = Some(addr.clone());

idx is never reassigned, so the DialResponse always carries addr_idx == 0 regardless of which address was actually tested.

The client maps the outcome back to an address using that index:

rust-libp2p/protocols/autonat/src/v2/client/handler/dial_request.rs

Lines 285 to 289 in 6a6814b

	let tested_address = req
	.addrs
	.get(res.addr_idx)
	.ok_or_else(|| io::Error::new(io::ErrorKind::InvalidInput, "address index out of bounds"))?
	.clone();

So when a DialRequest carries more than one address, the server dials the last one while the client attributes the result to the address at index 0.

Impact

DialRequest.addrs is a repeated field and addr_idx exists specifically to disambiguate which address was tested, so the wire format permits multi-address requests. The rust-libp2p client currently sends a single address per request (addrs: vec![addr]), so rust-to-rust is unaffected (the popped element is the only one, index 0). A spec-conformant client that sends multiple addresses would receive a reachability result attributed to the wrong address.

Separately, pop() selects the last (lowest-priority) address rather than the first.

Question

Is handling multi-address DialRequests in scope for the v2 server? If so the fix is small — either test addrs[0] and keep addr_idx = 0, or report the index of whichever address is tested. Happy to open a PR with a focused regression test once the preferred direction is confirmed.

Version

master (6a6814bb698b01c40d6ce08eb0f61281c909124f)