Add a baseline model for the bundled Charpy impact-energy data

[Krataios14]

FTQS already ingests the 78-row Charpy impact-energy sheet into assets/hea_impact_energy.csv, but the data is not used by the current qualification workflow. A small companion baseline would show what can honestly be learned from this asset without mixing impact energy with fracture toughness.

A useful first implementation would:

• Treat Charpy impact energy as its own target and keep it separate from KIC/KQ/J-derived values.
• Reuse the existing composition parsing and physical descriptors where they are applicable.
• Use a deterministic, group-aware evaluation so related alloy or source records do not leak across the split.
• Compare the chosen model with a simple constant or linear baseline.
• Report sample counts, held-out errors, uncertainty, and the limits imposed by a 78-row dataset; avoid making a coverage claim the split cannot support.
• Provide a reproducible command that writes a compact metrics artifact or report.
• Add focused tests for data loading, grouping, deterministic output, and target/unit handling.
• Document the method and its limitations in docs/METHODS.md and the README.

The existing fracture-toughness workflow and artifacts should remain unchanged by default. PLAN.md, src/ingest_fan2023.py, and tests/test_ingest_fan2023.py are good starting points.