Local object in Eval Api includes results for the whole graph even if running on a subgraph

Pagerank example:

    let out: HashMap<VID, f64> = runner.run(
        vec![Job::new(step1)],
        vec![Job::new(step2), Job::new(step3), Job::new(step4), step5],
        PageRankState::new(num_vertices),
        |g, _, _, local| {
            local
                .iter()
                .enumerate()
                .map(|(v_ref, score)| (v_ref.into(), score.score))
                .collect::<HashMap<_, _>>()
        },
        threads,
        iter_count,
        None,
        None,
    );

If this is run on a VertexSubgraph the local.iter() part still iterates over the whole graph, which might be unexpected. Might also need to be careful with the assumption that the position in the list is equal to the vertex id when subgraphs are involved.

Start with the Pagerank example and the Eval API callback using VertexSubgraph and local.iter(). Check how local results are constructed and whether vertex positions are treated as IDs. Done means subgraph evaluation returns only the intended vertices with correct identifiers, while full-graph behavior remains correct.