Common ScoreResult schema across all scoring adapters

[MaxGhenis]

Every scoring adapter currently returns a different dict shape (delta_gdp_bn rows vs changes_bn/changes_pct vs decile rows). Comparability across model classes is asserted in prose ("every model reports the same quantities") but not enforced by schema — so the site's central promise, the same reform through different model classes, side by side, has no programmatic object.

Contract — one ScoreResult (pydantic) for every scoring entry point:

• model (id + class: microsim / semi-structural / OLG-GE), country, reform (PolicyEngine parameter dict), horizon (steady-state | quarterly window | annual window)
• quantities: GDP, consumption, investment, government, revenue, debt — levels and deltas, currency-bn, with units and basis per quantity
• assumptions: list[str] (e.g. "pooled-age tax functions, single sector, steady state"), caveats: list[str] (e.g. the corp-tax boundary), uncertainty: optional bands where the model produces them (SVAR)
• Distributional block optional (microsim fills deciles/winners/losers; macro members leave it empty — until the Macro→micro feedback: economic-assumptions overlay for dynamic distributional impacts #11 feedback loop fills it)

Payoff: macromod compare --reform reform.json --models og,obr renders one table from N ScoreResults; the MCP returns them as a list. That table is the product moment.

Touches integration/src/macromod/core.py (adapters return the schema), cli.py (compare command), mcp_server.py, docs page. Pairs with #8/#9.