Expand the TransactionVerifier trait

[hackaugusto]

miden-node/block-producer/src/txqueue/mod.rs

Lines 53 to 59 in 1462d45

	#[async_trait]
	pub trait TransactionVerifier: Send + Sync + 'static {
	async fn verify_tx(
	&self,
	tx: SharedProvenTx,
	) -> Result<(), VerifyTxError>;
	}

Implementations of the above trait are responsible to validate transactions, this has to account for transactions is the store, as well as transactions currently in flight. Namely, when two transactions try to consume the same nullifier, only one can succeed.

The issue with the above traits are the following:

• The trait doesn't expose the complete API necessary to track in-flight transactions. This needs to be done with another trait, currently ApplyBlock, to callback this object and signal that an in-flight transaction has been committed to store, allowing the object to clean its internal state. While this works, it makes it harder to follow the code, since the interface is not "self sufficient".
• The trait doesn't have an intermediary state, between this transaction has been added to the queue, to this transaction has been added to a batch. It would be interesting to model such a state, because once fees are implemented, we would like to prioritize transactions not by arrival order, but by amount of fees paid.

[hackaugusto]

A sketch of a possible design is (pseudo code):

#[async_trait]
pub trait TransactionPool: Send + Sync + 'static {
    async fn tx_added(&self, tx: SharedProvenTx) -> Result<(), VerifyTxError>;
    async fn batch_created(&self, tx: Vec<SharedProvenTx>) -> Result<(), VerifyTxError>;
    async fn block_commited(&self, tx: Block) -> Result<(), VerifyTxError>;
}

We may want the above as a trait because of #23, so we can implement an in-memory and DB implementations (although we sqlite we can have an in-memory DB, so that is not super necessary, and we could have a plain implementation instead).

The design above makes sense to me since the lifecycle of the transaction changes after it touches multiple components. Which is kinda of modeled in the example above.

[hackaugusto]

A proposed flow is at #191 (comment)

[Mirko-von-Leipzig]

Rendered obsolete as the mempool now handles inflight tracking.