SSI state by-area recipient rows bind a mixed universe (~115k state-supplement-only gap)

Residual from the #507/#508 adjudication, flagged for separate adjudication in the #508 close-out (now closed, so tracking here).

The by-area STATE recipient rows (ssa_supplement.ssi_recipients, role ssi_recipients) still bind engine ssi — the federal payment — while their universe is federally administered, which includes ~114,977 state-supplement-only recipients nationally (7,404,820 − 7,289,843, December 2024). The national all_areas_total row was re-scoped non-binding for exactly this mismatch (#524), but the 51 state rows remain the only state-grain SSI caseload signal, so they kept binding with the residual documented at the compile site (_ssa_ssi_reference_from_fact in fiscal_targets.py).

Closing it needs one of:

1. a state-level federal-payment recipient source (the SSA SSI Monthly state tables, if the federal-payment split exists at state grain), or
2. a modeled state-supplement concept, letting the federally-administered rows bind a combined federal+supplement quantity.

Until then the solve carries a ~1.6% wrong-universe pull distributed across states. Relates #508, #524, #549.

Start at _ssa_ssi_reference_from_fact in fiscal_targets.py and inspect how the ssa_supplement.ssi_recipients state rows bind to engine ssi. Review the close-out context in #508 and the re-scoping in #524, then determine whether a state-level federal-payment source exists or a modeled combined concept is needed. Done means the state rows bind a matching universe without the documented residual.