Collators don't clear connection requests until they produce another collation

[rphmeier]

polkadot/node/network/collator-protocol/src/collator_side/mod.rs

Lines 518 to 533 in c1a8115

	/// Issue a connection request to a set of validators and
	/// revoke the previous connection request.
	async fn connect_to_validators<Context>(ctx: &mut Context, validator_ids: Vec<AuthorityDiscoveryId>)
	where
	Context: SubsystemContext<Message = CollatorProtocolMessage>,
	Context: overseer::SubsystemContext<Message = CollatorProtocolMessage>,
	{
	// ignore address resolution failure
	// will reissue a new request on new collation
	let (failed, _) = oneshot::channel();
	ctx.send_message(NetworkBridgeMessage::ConnectToValidators {
	validator_ids,
	peer_set: PeerSet::Collation,
	failed,
	})
	.await;

is only invoked here, after a collation has been sent to the collator-protocol subsystem

polkadot/node/network/collator-protocol/src/collator_side/mod.rs

Lines 392 to 404 in c1a8115

	tracing::debug!(
	target: LOG_TARGET,
	para_id = %id,
	relay_parent = %relay_parent,
	candidate_hash = ?receipt.hash(),
	pov_hash = ?pov.hash(),
	core = ?our_core,
	?current_validators,
	"Accepted collation, connecting to validators."
	);
	// Issue a discovery request for the validators of the current group:
	connect_to_validators(ctx, current_validators.validators.into_iter().collect()).await;

If there is no collation for some time, an outdated connection request will remain in-memory.

[slumber]

Better to sync it with #5226 since it changes the connection logic