feat(eeio): derive y_nab from Adom and scaled_q

[bl-young]

cc:
Closes: #456

What changed? Why?

y_nab_USA for the Cornerstone full model (2025_usa_cornerstone_full_model) is derived from the same derive_cornerstone_Aq_scaled() object that produces snapshotted scaled_q_USA and Adom_USA, via backcompute_y_from_q_and_Aq(Adom, scaled_q). That enforces IO row balance q = Adom @ diag(q) + y_nab and domestic Leontief consistency q ≈ L_dom @ y_nab.

The prior path built y_nab from BEA Y/trade disaggregation and inflation — parallel to the A/q pipeline and inconsistent with snapshotted scaled_q (323 commodity sectors failed q ≈ L·y at 1% tolerance on the domestic branch).

derive_cornerstone_y_nab — replaces the BEA disaggregation/inflation body with a three-line backcompute from derive_cornerstone_Aq_scaled(). Negative y_nab values are retained; handle_negative_vector_values breaks the domestic Leontief identity because L_dom is dense (24 clipped sectors caused 26 test failures during development).

test_compare_output_and_L_y — adds a passing Cornerstone domestic case (Commodity, use_domestic=True). Total Leontief cases (Cornerstone and CEDA) remain xfail; Cornerstone total still pairs scaled Adom + Aimp with BEA-derived y from derive_cornerstone_Y_and_trade_scaled.

Unchanged: scaled_q, Adom, ytot, ydom, yimp, exports derivation paths. CEDA v7 legacy path. derive_y_for_national_accounting_balance_usa router (Cornerstone branch still delegates to derive_cornerstone_y_nab).

Follow-ups (not in this branch)

• Snapshot bump — y_nab_USA values change; test_y_nab_usa_snapshot fails until Phase A regeneration (GCS upload, .SNAPSHOT_KEY, releases.py, usa_config.py Literal). Required before merge if integration tests gate on snapshots.
• Refine q estimate #457 — 457-refine-q-estimate refines scaled_q via authoritative-x V scaling. y_nab couples to derive_cornerstone_Aq_scaled(), so refined q propagates automatically once wired into that path. Coordinate stack order if both land near the same snapshot cut.
• Total L·y and Y snapshots — reconcile ytot/ydom/yimp/exports with IO-balanced y_nab, or adjust MRIO multi-region Y assembly (ydom + exports vs y_nab_USA parquet).

Testing

Domestic Cornerstone Leontief identity (passes, 0 failing sectors at 1% rtol):

uv run pytest "bedrock/utils/validation/__tests__/test_eeio_diagnostics.py::test_compare_output_and_L_y[Commodity-True-cornerstone]" -m eeio_integration -v

Full diagnostics module (domestic passes; total cases xfail as expected):

uv run pytest bedrock/utils/validation/__tests__/test_eeio_diagnostics.py -k test_compare_output_and_L_y -m eeio_integration -v

Snapshot integration (fails until y_nab_USA snapshot regeneration):

uv run pytest bedrock/transform/__tests__/test_usa.py::test_y_nab_usa_snapshot -m eeio_integration -v