The current built-in node crashes with a Couldn't find coin for input error when a block containing txs gets reorged in some situations.

Here's what happened (with the bug-relevant bits bold):

1. Connected block A, which contained tx T
2. Received block B, which also contained T
3. Avalanche selected block B to be finalized
4. Node disconnected block A (Chronik didn't receive this yet)
5. Node added T back to the mempool (due to the re-org)
6. Node connected block B, removing T and marking its coins as spent
7. Only now Chronik disconnects block A
8. Chronik adds T back to the mempool (due to the re-org), however, FindCoins can't find the coins anymore, as they've been spent in step 6.

To fix this, we thread the spent_coins from the node to the validation interface queue, by adding spent_coins to TransactionAddedToMempool. This way Chronik is guaranteed to have the spent coins ready without having to query them from the node.

We use a std::shared_ptr<const std::vector<Coin>>, to ensure coins aren't freed and reused before they're used by the indexer. The interfaces take an owned shared_ptr, to make sure the memory isn't freed too early. In the (relatively far) future, we might move spent_coins into CTransaction, which would manage the memory for us.

List of changes:

1. Add (Rust) CCoin to the FFI pointing to C++'s Coin.
2. Change bridge_tx to take a vector of spent coins instead of ChronikBridge (and fix the bug).
3. Thread the spent coins to handle_tx_added_to_mempool and feed them to bridge_tx.

Depends on D14386.

ninja check-functional