[Feat] make supercircuit test applying l2trace

[noel2004]

In scroll feature, this PR directly connect the l2trace generated by l2geth_util with supercircuit

It has induced some utilities inside scroll-prover before and some refactoring on the zktrie crate.

With this PR the circuit builder in bus-mapping now support a building route from l2 trace

Notice it contains some breaking changes for scroll-prover

[noel2004]

Has to disabled the doc-test in busmapping: #782

[lispc]

log::debug!(
            "building zktrie state for block {:?}, old root {}",
            l2_trace.header.number,
            hex::encode(old_root),
        );

        let mpt_init_state = ZktrieState::from_trace_with_additional(
            old_root,
            Self::collect_account_proofs(&l2_trace.storage_trace),
            Self::collect_storage_proofs(&l2_trace.storage_trace),
            l2_trace
                .storage_trace
                .deletion_proofs
                .iter()
                .map(Bytes::as_ref),
        )
        .unwrap();

        log::debug!(
            "building partial statedb done, root {}",
            hex::encode(mpt_init_state.root())
        );

these lines cost ~35ms. Too much. This is the reason why ccc becomes much slower. I will refactor this.

inside update_nodes_from_proofs, this line zk_db.add_node_bytes(bytes).unwrap() is slow. Some poseidon computation is applied here.

one solution is like my previous code, use light_mode to skip filling zktrie mem db. Another solution is to provide hash results inside trace (if it is free), and avoid doing poseidon again in Rust.