SDLT calibration target includes corporate SDLT, which policyengine-uk adds separately #564

Problem

targets/sources/obr.py maps the OBR's total stamp duty land tax line (receipts Table 3.8, all buyers) to the household variable stamp_duty_land_tax. _compute_simple_gbp then sums that variable over households. This is the same code in the 1.58.0 release (obr.py line 264).

policyengine-uk adds SDLT paid by companies separately: corporate_sdlt is shareholding × HMRC's receipts from non-natural persons, and it sits in gov_tax and household_tax alongside expected_sdlt. So calibration pushes household SDLT up to a total that already includes the corporate share, and the model then adds the corporate share again.

HMRC's buyer-type tables (Annual Stamp Tax Statistics 2024 to 2025, Tables 8a and 8b) put non-natural persons at £4,490m of £13,885m in 2024-25: £1,230m residential and £3,260m non-residential, or 32.3%. The OBR's 2024-25 outturn for SDLT by property type (receipts Table 3.7) equals HMRC's all-buyer totals exactly: £10,380m residential and £3,505m non-residential.

Measured on 1.58.0

enhanced_frs_2024_25.h5, sha256 5ba399ea…d960b6, policyengine-uk with PolicyEngine/policyengine-uk#2322, which projects corporate SDLT with the OBR forecast. Figures are £bn.

Before #2322, corporate SDLT was held at £3.13bn and the 2026-27 total was £20.44bn.

Fix

Target stamp_duty_land_tax at the OBR total less policyengine-uk's corporate statistic for the same year: gov.hmrc.stamp_duty.statistics.{residential,non_residential}.corporate.revenue, read from the installed model.

After #2322 that statistic is HMRC's 2024-25 non-natural-person share of each property type, applied to the OBR's forecast for that type. The target is then individuals' SDLT on the OBR's forecast: £18.15bn − £5.98bn = £12.17bn in 2026-27.

tests/test_property_purchased.py bounds weighted expected_sdlt against total receipts (£5–50bn), and should compare against the household-only figure. #556's household non-residential targets can read the same statistics; #2322 updates them to 2024-25 (£245m, 25,000 transactions, £7.64bn).

This belongs in the batched data release.