fix(network): stop demoting healthy peers on simultaneous-dial dedup

`deregister_stream` recorded `node_db.note_failure` for every expired
session it removed, but the simultaneous-dial dedup kill passes
`op = None` (since #3436 / 4ab4a79) precisely to avoid recording a
failure — so the dropped redundant connection demoted the healthy peer
anyway. The same blanket call also dragged a peer back down after it had
already reconnected.

Remove the blanket `note_failure` from `deregister_stream`. Peer
reputation on disconnect is already recorded by the kill path via its
`UpdateNodeOperation` (note_failure / demote / set_blacklisted for remote
failures), and `set_expired` is only ever set by the two kill functions,
so no genuine failure recording is lost.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: peilun-conflux

crates/network/src/service.rs

@@ -1814,12 +1814,8 @@ impl IoHandler<NetworkIoMessage> for NetworkServiceInner {
                     if sess.expired() {
                         sess.deregister_socket(poll_registry)
                             .expect("Error deregistering socket");
-                        if let Some(node_id) = sess.id() {
-                            self.node_db.write().note_failure(
-                                node_id, true,  /* by_connection */
-                                false, /* trusted_only */
-                            );
-                        }
+                        // Reputation on disconnect is recorded by the kill
+                        // path (its `UpdateNodeOperation`), not here.
                         self.sessions.remove(&sess);
                         debug!("Removed session: {:?}", *sess);
                     }