Impute US humanitarian immigration statuses to cited stocks (H.R.1 channels)

[MaxGhenis]

Fixes #767.

Builds humanitarian and temporary-protection immigration statuses into the US Microcosm source stage and stacked production pipeline, so downstream eligibility rules—including the H.R.1 Medicaid, ACA, and SNAP channels—receive nonzero, source-backed populations.

What changes

Source-aware immigration evidence

• Keeps citizenship measured rather than imputed.
• Normalizes native ASEC evidence (PRCITSHP, PENATVTY, PEINUSYR) and native ACS evidence (CIT, POBP, YOEP) without coarsening exact ACS arrival years.
• Requires valid foreign-born ACS arrival evidence for CIT=4/5 rows and applies exact country-specific TPS continuous-residence cutoffs.
• Preserves the paired ssn_card_type / immigration_status_str contract, including deterministic Cuban/Haitian entrant and DACA reconstruction.

Cited humanitarian controls

The manifest carries one cited control for each draw:

• parole: Afghanistan 73,566; Ukraine 158,000; Nicaragua 93,070; Venezuela 117,330;
• refugees: 160,000, based on the trailing FY2023–FY2024 admissions reported by DHS OHSS;
• asylees: 155,000, the rounded FY2022–FY2024 grant total reported by DHS OHSS;
• TPS: Venezuela, El Salvador, Honduras, Nicaragua, Nepal, and other designated countries from CRS RS20844;
• deportation withheld: an explicit cited zero because no defensible stock estimate is published at CPS resolution.

The four parole controls use DHS/CBP administrative publications linked directly in source_stages.json. Cuba/Haiti-born people, the DACA cohort, and overlapping Ukraine/Afghanistan TPS cohorts retain the more appropriate statutory labels described in the manifest notes.

Pew residual-estimate semantics

• Uses Pew's current, coherent 2023 pair: 14.0 million unauthorized immigrants and 9.7 million in the civilian labor force. Pew's 2025 series supersedes the former 11.0 million / 8.3 million estimates used on this branch.
• Applies the controls to the modeled Pew universe rather than only the engine's narrower UNDOCUMENTED enum. DACA, parole, TPS, deportation-withheld, and residual Cuban/Haitian EAD rows remain included; refugees, granted asylees, and assumed-documented Cuban/Haitian entrants remain excluded.
• Restores measured A_LFSR from the same SHA-pinned official Census ASEC person archives by exact (source_year, PERIDNUM) join. The 9.7 million margin uses age 16+ and A_LFSR codes 1–4 (working, job but absent, looking, or layoff), not prior-year wage/self-employment income.
• Keeps the 14.0 million population total as a coarse release diagnostic rather than forcing labels to it. The worker and student spill margins remain the only forced totals.
• Does not add Pew to Chronicle; Chronicle remains limited to government facts.

Stacked transfer and reconciliation

• Adds the paired immigration outputs and their required source predictors to the canonical late-producer registry.
• Treats ASEC rows as immutable evidence and allows constrained reconciliation only on paired native ACS recipient rows.
• Reconciles each draw to the residual target after immutable ASEC mass, in manifest order, using stable lineage-based seeded selection.
• Scales controls when building sampled pools so a rung represents the same full-population targets.
• Persists QRF-pattern evidence and one canonical reconciliation receipt on both paired outputs.
• Replays live weighted populations and the exact deterministic selected-row identities at checkpoint, H5 load, final transition-authority, and publication boundaries. Re-digesting a tampered H5 file is insufficient to forge this proof.

The merge with current main preserves the upstream pregnancy structural-transfer proof alongside the immigration reconciliation proof; both receipts are required where applicable.

Calibration and release gates

• Adds 12 positive government humanitarian stock targets to the fiscal calibration registry; the explicit-zero withholding draw remains a gate constraint rather than a positive calibration row.
• Materializes the source-aware person masks to household calibration columns for both dense and sparse calibration.
• Rechecks immigration composition on final calibrated export weights, including the modeled Pew population diagnostic.
• Makes immigration_composition part of the exact canonical stacked terminal-gate set, including checkpoint/H5 authentication and failed-publication receipts.
• Retains four release-blocking reform probes for the Medicaid, ACA lawful-presence, below-FPL PTC, and SNAP eligibility channels.

Validation

• Spec-engine coverage: 42,538 / 42,538 configuration fields and 41 / 41 inventory checks.
• Merged US spec: ae83c7a32a4a2970070f71b95aa7de771a56ab435f71004c11ee2e6b181fe07b.
• All 17 affected local test suites pass, including immigration, ASEC sidecar/checkpoint, build-path, spec compilation/coverage, late-producer DAG, multispine, and bundle-contract tests.
• Bundle generation and coverage checks pass; Ruff lint, changed-file format checks, and git diff --check pass.
• A real-archive smoke test covers all 432,523 records across income years 2022–2024 with only the official A_LFSR domain {0,1,2,3,4,7}.
• Independent review found no remaining actionable issues.

Deliberately out of scope

• US: store years_since_us_entry from PEINUSYR (blocked on engine ICHIA option) #776 — store years_since_us_entry once the engine exposes the remaining duration consumers.
• Verification/redetermination churn flag surface (option 2 of #767, generalized) #777 — verification/redetermination churn flags.
• Add the CHIP §71109 reform-coverage probe when the engine encodes it (pe-us#9346) #778 — add a CHIP probe once the corresponding engine channel is available.

No certified release artifact is changed by this PR; the controls activate through the next US build lane and its authenticated release gates.