Add federal cost-allowability analysis component set and workflow

[ProfessorPolymorphic]

Summary

• Adds a 13-component set for analyzing the allowability of federal-grant expenses under the Uniform Guidance (2 CFR 200), decomposed so each component maps to a discrete regulatory requirement, plus the 3-step cost-allowability-analysis Vandalizer workflow that combines them.
• Catches expenses tied to IRB / IACUC / IBC-regulated activities via a regulated-activity-classifier detection layer and a protocol-approval-allowability-check — a regulatory regime distinct from 2 CFR 200.407 federal prior approval.
• Federal-generic (2 CFR 200), not sponsor-specific; the existing NSF-specific nsf-expense-allowability-check is untouched and coexists as the fast single-shot reviewer.

Components (all v0.1.0, status experimental)

Stage	Components
Normalize & detect	expense-transaction-extraction, award-allowability-terms-extraction, regulated-activity-classifier
Allowability checks	cost-period-check, cost-reasonableness-check, cost-allocability-check, cost-consistent-treatment-check, cost-award-terms-conformance-check, cost-prior-approval-check, cost-documentation-check, protocol-approval-allowability-check
Selected items	cost-selected-item-check
Synthesis	cost-allowability-determination

The 9 checks emit a shared structured finding contract; the determination emits Markdown. All 13 are registered in component_catalog_overrides.yaml; component_catalog.json, the Vandalizer export, and the docs nav are regenerated.

Validation

• python3 scripts/build_component_catalog.py --check — up to date
• python3 scripts/build_vandalizer_workflows.py --check — all exports up to date
• python3 .github/scripts/lint_components.py — 0 errors (2 pre-existing warnings unrelated to this change)
• python3 -m mkdocs build --strict — clean

Test plan

• Upload workflows/cost-allowability-analysis/cost-allowability-analysis.vandalizer.json into Vandalizer and run an end-to-end expense review
• Confirm Step 3 (synthesis) receives the Step 2 findings as expected under the linear step model
• Add golden eval cases per component (evals/cases/) to graduate components from experimental

🤖 Generated with Claude Code