Fix1

[Idleness76]

What I changed

• Removed pre-filtering in App::invoke so the App passes the full frontier to the Scheduler and relies on the Scheduler’s decision.
• Made the Scheduler return strongly-typed node IDs (NodeKind) instead of strings to avoid formatting/parsing churn.
• Updated demos/tests to the new API and behavior.

scheduler.rs; StepRunResult now uses:

1. ran_nodes: Vec
2. skipped_nodes: Vec
3. outputs: Vec<(NodeKind, NodePartial)>
4. superstep still partitions the frontier and executes nodes with bounded concurrency.
5. should_run/record_seen behavior unchanged; only usage consolidated inside superstep.

app.rs

1. Removed the .filter(|n| scheduler.should_run(...)) on the frontier.
2. Uses step_result.ran_nodes and step_result.outputs to build node_partials and compute the next frontier.