Design a durable program-statistics mapping for changing country packages #326

Follow-up to Vahid's JOSS PR: #264

The immediate US breakage is tracked separately in #325. This issue is for the broader design problem: program-statistics analysis currently depends on hard-coded country program variable names and output assumptions, but country packages can rename, add, remove, or re-scope program variables over time.

The immediate fix can patch the current US mappings, validate them earlier, and improve the error message. That does not solve the underlying fragility. A durable design should make program-statistics mappings explicit, version-aware, and tested against the variables that are actually materialized into output datasets.

Potential directions:

• Maintain country-specific program-statistics metadata with explicit display names, concrete variable names, entities, tax/benefit polarity, and supported periods/package versions.
• Validate mappings before expensive simulations run, including both model-variable existence and output materialization.
• Prefer metadata or registry-driven program definitions where the country package exposes stable program concepts.
• Add drift-catching tests that load the current country model metadata and assert every configured program-stat variable can be aggregated from a small output fixture.
• Decide how to handle semantically ambiguous programs, especially cases like Medicare where the intended concept could be benefits, enrollment, payroll tax, or spending.

This should be treated as separate from the immediate repair PR, which is only intended to restore the currently broken US economic-impact path and make future failures easier to diagnose.