kad: Query::next_action() message is ignored by the kademlia handler

[lexnv]

Query::next_action() method returns a QueryAction that contains a raw kademlia message.
For the FindNode query type it is the FIND_NODE message, for GetRecords query type it is the GET_VALUE message.

litep2p/src/protocol/libp2p/kademlia/query/find_node.rs

Line 232 in e3a22d5

pub fn next_action(&mut self) -> Option<QueryAction> {

The message is ignored by the kademlia handler on_query_action method:

litep2p/src/protocol/libp2p/kademlia/mod.rs

Lines 787 to 791 in e3a22d5

	while let Some(action) = self.engine.next_action() {
	if let Err((query, peer)) = self.on_query_action(action).await {
	self.disconnect_peer(peer, Some(query)).await;
	}
	}

The following lines are ignoring the provided message, and sending the FindNode request regardless of the message provided:

litep2p/src/protocol/libp2p/kademlia/mod.rs

Lines 700 to 707 in e3a22d5

	QueryAction::SendMessage { query, peer, .. } => {
	if self
	.open_substream_or_dial(peer, PeerAction::SendFindNode(query), Some(query))
	.is_err()
	{
	// Announce the error to the query engine.
	self.engine.register_response_failure(query, peer);
	}

The intention behind this behavior was to advance the discoverability of the network. For example, GetRecords query must first discover the network, then publish a GET_VALUE request.

Consider untangling the query message with the kademlia handler and to allow queries to decide which message should be sent.
We can further look into optimizing the number of messages sent from the query engine to the kademlia handler (since some of them may be redundant as in this case)