smc: revise defer Unsubscribe comment to match actual lifecycle

Per Copilot — the previous comment claimed parity with the
SSE-stream-failure path, but in practice PeerConnectionEvents
blocks until ctx cancellation and there's no retry loop wrapping
listenPeerConnectionEvents, so the defer effectively runs at
process shutdown. Rewrote to spell that out while still calling
out why defer is the right shape (future-proofing against early
returns / retry wrappers).

No code change.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Author: myleshorton

lantern-core/core.go

@@ -425,12 +425,15 @@ func (lc *LanternCore) listenPeerConnectionEvents() {
 	// hit here today. Worth revisiting if Unbounded ever moves out of
 	// process.
 	//
-	// defer Unsubscribe so cleanup runs on EITHER path that exits
-	// this function: ctx cancel (normal) OR PeerConnectionEvents
-	// returning an error (unexpected stream exit). The earlier
-	// ctx-watching goroutine missed the error path and leaked both
-	// the goroutine and the subscription whenever the SSE stream
-	// died while ctx was still live.
+	// defer Unsubscribe so cleanup runs whenever this function
+	// returns. In practice that's process shutdown — the function
+	// is started once from initialize(), the PeerConnectionEvents
+	// call below blocks until ctx cancellation, and there's no
+	// retry loop wrapping it. But if PeerConnectionEvents ever did
+	// exit early on stream error (or a future caller adds retries),
+	// defer keeps the subscription lifetime bound to this function's
+	// scope rather than leaking until the next reinitialization —
+	// which the previous ctx-watching goroutine implicitly relied on.
 	unbSub := events.Subscribe(func(evt unbounded.ConnectionEvent) {
 		jsonBytes, err := json.Marshal(map[string]any{
 			"state":     evt.State,