Add generic metadata field to `TxNode`

[luisschwab]

In order to integrate different metaprotocols with BDK, such as Utreexo and Silent Payments, an additional metadata field is needed in bdk_chain, under TxNode. e.g.:

/// A transaction node in the [`TxGraph`].
#[derive(Clone, Debug, PartialEq, Eq, PartialOrd, Ord)]
pub struct TxNode<'a, T, A, M> {
    /// Txid of the transaction.
    pub txid: Txid,
    /// A partial or full representation of the transaction.
    pub tx: T,
    /// The blocks that the transaction is "anchored" in.
    pub anchors: &'a BTreeSet<A>,
    /// The first-seen unix timestamp of the transaction as unconfirmed.
    pub first_seen: Option<u64>,
    /// The last-seen unix timestamp of the transaction as unconfirmed.
    pub last_seen: Option<u64>,
    /// The last-evicted-from-mempool unix timestamp of the transaction.
    pub last_evicted: Option<u64>,
    /// Transaction Metadata
    pub metadata: Option<M>,
}

[luisschwab]

cc @nymius

[nymius]

I thinking along the same lines. What I'm concerned about is the interaction of these new nodes with the Indexer and IndexedTxGraph.
The idea of dependency inversion is starting to click on me, like in CanonicalizationTask, one of the first ideas that comes to my mind is to allow the indexer trait to have an extra MetadataProvider to request this data at the moment of inserting the transaction in the TxGraph.
But then, the Indexer would have a mixed logic, where you push Tx and pull data from other resource (I have to think more on this).
Also, maybe a bound to the M trait would ensure the persistence details are on the side of the Metadata implementer.

[luisschwab]

We had a few discussions about this on the Utreexo Summit and arrived at the conclusion that this will not be necessary for bdk_floresta.

Since the inclusions proofs are dynamic–the proof for TxOut A at height X is different of the proof for the same TxOut at height X+2–a lot of writing to TxGraph would be necessary. Instead, we defer the job of proof storage and caching to the Floresta node. At spend time, we request the proof from it, attach it to the transaction, and then broadcast it.

[nymius]

After talking with @evanlinjin , I tracked back the issue to the original conflicting point for silentpayments, the IndexedTxGraph and the Indexer trait.
The last one is not general enough to allow the construction of an IndexedTxGraph tracking spks for a BIP 352 compatible wallet. Instead of modifying TxNode to host the tweaks, a more acceptable change would be to modify this Indexer trait or change the interaction between TxGraph and Index different, so extra metadata from transactions can be indexed atomically.