X-chain SDK gossip

[joshua-kim]

Why this should be merged

This PR introduces p2p SDK based tx gossip to the X-chain. This includes pull and push gossip of transactions (under a new message format) while maintaining compatibility with the prior push gossip mechanism.

How this works

• Makes *txs.Tx implement the Gossipable interface
• Creates a gossip mempool that wraps the prior mempool with a bloom filter and (optional) verification checks.
• Separates the AVM config into it's own file to better handle the new networking based configs
• Modifies the existing network struct to pass all AppRequest, AppResponse, and AppRequestFailed messages to the p2p SDK network implementation.
• Modifies the existing network struct to pass AppGossip messages that can't be parsed with the prior format to the p2p SDK network implementation.
• Deprecates the prior push gossip mechanism.

How this was tested

• CI
• Tested on fuji
• Tested on a local network while explicitly dropping the legacy message format
• Tested on a local network while explicitly dropping the new message format

[dhrubabasu]

Should we check if tx is already in the mempool?

[StephenButtolph]

AddVerified is expected to error if the tx is already in the mempool (which is what happens here)

[patrick-ogrady]

Should we try to better batch these gossip invocations to reduce the number of messages we push (no more than one every x ms)? Or is that handled in the SDK?

[patrick-ogrady]

AFAICT we would batch if we didn't invoke manually:

avalanchego/network/p2p/gossip/gossip.go

Lines 331 to 347 in ee10054

	// Every calls [Gossip] every [frequency] amount of time.
	func Every(ctx context.Context, log logging.Logger, gossiper Gossiper, frequency time.Duration) {
	ticker := time.NewTicker(frequency)
	defer ticker.Stop()
	for {
	select {
	case <-ticker.C:
	if err := gossiper.Gossip(ctx); err != nil {
	log.Warn("failed to gossip", zap.Error(err))
	}
	case <-ctx.Done():
	log.Debug("shutting down gossip")
	return
	}
	}
	}

[StephenButtolph]

Yeah - this is kind of an independent suggestion to this PR (as this is the same behavior as before)... But we could play around with batching

[patrick-ogrady]

Would it be better to just key off of support based on the handshake?

[patrick-ogrady]

I think we already get version in the Connected message, so it may not be that bad but could be wrong.

[StephenButtolph]

We use AppGossip for this, so we don't have the ability to select who exactly we are sending the messages to for push gossip. We could filter the pull gossip to only send it to newer nodes if we cared to optimize this during the time between this release and the network upgrade.