Floor the uk extra at policyengine-uk>=2.92 for the pension-wealth split (#750)

[juaristi22]

Closes #750. This is the closing increment: the split machinery merged via #755, the model side merged as PolicyEngine/policyengine-uk#1824 and shipped as policyengine-uk 2.92.0, the lever has its first calibrated measurement (v11/v12: UC initial support 3,128,938 → 4,242,978 benefit units, caseload −25.6% → −20.8%, clean isolation — readout on the issue), and this floor bump is the last microcosm-side item on the post-merge checklist. The remaining checklist items graduate on closure: the uk-data mirror decision is Max's call with its ordering already tracked on #796 (B4: mirror → efrs-post-calibration re-pin), the wealth re-draw residual is now owned by #809 (mechanism 1) with the #145 predictor-quality revisit, and the composite mass audit, cash-ISA gap, and current_db_pension_wealth follow-ups move to the #796 deferral ledger.

What this PR does

The #755 WAS split ships spines whose corporate_wealth excludes private pension wealth, and policyengine-uk 2.92.0 is the first release carrying private_pension_wealth as an explicit capital disregard plus corporate_sector_wealth for the shareholding / corporate-land keys. This floors the uk extra at >=2.92 and re-locks (single-package upgrade, 2.89.0 → 2.92.1), so a national build or calibration on a split dataset can never resolve an engine that drops the column.

Context: the v11/v12 assessment runs on spine-j resolved 2.89.0 under the old floor. Measured and found insensitive for the bound register surface — ons.land.* initial estimates are byte-identical across the split (corporate land does not flow through corporate_wealth) and the UC capital test reads corporate_wealth alone under both versions — but the certified cut must not repeat the coincidence.

🤖 Generated with Claude Code