Harden DFT transferability evidence

[benjipeng]

Summary

• persist runtime and implementation identities consistently across EOS artifacts
• fail closed on SCF point-group reductions that lack density reconstruction or a matched full-mesh oracle
• replace the unsafe MgO q10 feasibility route with a full-mesh future route
• add and run a Fe q16 full-versus-reduced SCF oracle, retaining its failed energy and moment gates
• reconcile the Phase 7 contract, roadmap, and material evidence documentation

Verification

• 33 focused DFT tests passed
• ruff check src passed
• API documentation generation passed
• canonical site documentation sync check passed
• current Fe q16 oracle: 0.0056218 Ha/atom free-energy difference and 0.07977 Bohr-magneton-per-atom moment difference; both correctly fail their locked gates