fix: populate localAddress on connect handler Socket

[fhanau]

Clone of #6647 (already approved) – see that PR for context.

ServiceWorkerGlobalScope::connect was dropping the CONNECT authority when constructing the ingress Socket, causing both socket.opened.remoteAddress and socket.opened.localAddress to resolve to undefined on the connect handler side. Plumb a localAddress parameter through Socket/setupSocket and forward the host parameter into it so JS callers see the exact authority string that was passed to fetcher.connect(...).

From the handler's perspective the CONNECT authority is the local address on this side of the tunnel (the address the peer asked to connect to), not the remote address, so populate localAddress rather than remoteAddress. Outbound sockets continue to populate remoteAddress as before; localAddress remains empty for outbound since we have no useful value to expose.

Extend the existing connect-handler-test with a localAddressViaServiceBinding case that asserts strict equality between the authority passed to fetcher.connect(...) and the localAddress observed on the server side.

[ask-bonk]

I've now thoroughly reviewed all the changed files. Let me assess:

Summary of changes:

1. Added a localAddress parameter to Socket constructor, setupSocket, threaded through startTls and both handleProxyStatus overloads.
2. In ServiceWorkerGlobalScope::connect, forwards the host (CONNECT authority) as localAddress to the setupSocket call.
3. Outbound paths (connectImplNoOutputLock, Hyperdrive::connect) pass kj::none for localAddress — no behavior change.
4. Test coverage via service-binding round-trip test asserting strict equality.
5. Tail worker test adjustments for the new test case.

Review findings:

1. Correctness: The plumbing is consistent — every call site of setupSocket passes localAddress, every handleProxyStatus and startTls path propagates it. The kj::mv(host) in global-scope.c++ is safe since host is a kj::String parameter (owned).

2. Backward compatibility: The localAddress field previously always resolved to undefined in JS. Now it resolves to a string for inbound connect-handler sockets. The connect handler itself requires the experimental compat flag, so the blast radius is minimal. No compat flag needed.

3. Memory/GC safety: localAddress is kj::Maybe<kj::String> (C++ type, not JS heap), so it correctly doesn't need visitForGc. It also isn't in visitForMemoryInfo, but neither is remoteAddress — pre-existing oversight.

4. API design: Using localAddress rather than remoteAddress for the inbound handler is semantically correct — from the handler's perspective, the CONNECT authority is the address the peer asked to connect to on this end.

5. Test quality: The test exercises the service-binding path specifically (which forwards the authority verbatim) and asserts strict equality. Good red-green verification in the PR description.

I don't see any actionable issues.

LGTM

github run