Schema + viewer: per-node artifact/code pointers

[EYH0602]

Background

Gap analysis vs the deployed ara-hub viewer (https://www.agenticresearch.sh/ara/AmberLJC/ara-paperbench/artifacts/speedrun/nanogpt-speedrun). This is a schema-widening task (parser + manifest + viewer), not viewer-only.

Problem

The hub renders per-node artifact pointers — name, pointer (path), what — linking each step to the concrete code/config/heuristic it produced (e.g. "muon_optimizer.py — src/execution/muon_optimizer.py — Muon optimizer implementation: Newton-Schulz orthogonalization + Nesterov momentum").

Our Node struct (crates/ara-core/src/manifest.rs:255-281) has no artifact field, so this grounding link between exploration steps and physical artifacts never reaches the viewer.

Proposal

• Widen the node schema with an optional artifacts: list (name, pointer, what).
• Render as a "artifacts" block in the detail pane: mono path + name + one-line description, in published hub block order (after result, before sources).

Acceptance

• Node-level artifact pointers parse and appear in the manifest
• Detail pane renders an artifacts block when present, omitted when empty
• Parser + viewer unit tests