server: fix panic if heartbeat reset happens for GC'd node (#23383)

When setting up the timer for heartbeat invalidation, there's no control that
allows us to remove that timer when the node is GC'd. If the GC window is narrow
enough, it's possible to GC a node that has a waiting heartbeat timer. In this
case, we hit a bug where querying for the node returns `nil` and this is
incorrectly handled when checking for disconnect/reconnect state. Fix this bug
by correctly handling a `nil` node and allowing the `Node.Update` RPC to fire
normally (which then errors correctly).

Fixes: https://github.com/hashicorp/nomad/issues/23376
Ref: https://hashicorp.atlassian.net/browse/NET-10109

.changelog/23383.txt

@@ -0,0 +1,3 @@
+```release-note:bug
+server: Fixed a bug where expiring heartbeats for garbage collected nodes could panic the server
+```

nomad/heartbeat.go

@@ -183,6 +183,10 @@ func (h *nodeHeartbeater) disconnectState(id string) (bool, bool) {
 		h.logger.Error("error retrieving node by id", "error", err)
 		return false, false
 	}
+	if node == nil {
+		h.logger.Error("node not found", "node_id", id)
+		return false, false
+	}
 
 	// Exit if the node is already down or just initializing.
 	if node.Status == structs.NodeStatusDown || node.Status == structs.NodeStatusInit {