The `aggregate_balances_changes` was keeping a map of token -> balance_change per transaction. Therefore, if a transaction was causing a balance change for the same token but on differents components we would only keep the update for the last component updated and drop the others.
In case we get negative balances, which happens sometimes e.g. in balancer and seems to be due to some rounding issues within the protocol, we simply clip the balance at 0 so we never emit negative balances as tycho-indexer the upstream system will interpret the balance bytes as unsigned integers.
