Run explicit heterogeneity, mechanism, placebo, and alternative-specification analyses from a baseline econometric result.

Use this after a baseline model already exists and only when the user has explicitly named the subgroup or mechanism variables to test.

Behavior:
- Read baseline `results.json`, `diagnostics.json`, and `numeric_snapshot.json` first.
- Refuse to continue if the baseline already has blocking QA or diagnostic failures.
- Create one structured result directory per specification.
- Save unified summaries, narratives, and a publication-style comparison table.
- Every narrative claim must be grounded in each specification's own structured outputs.
