Fix silently-dropped OBR receipts targets (sheet 3.9 is APD, not current receipts)

[vahid-ahmadi]

Problem

_parse_receipts reads the cash-basis receipts rows from a hard-coded worksheet "3.9":

ws39 = wb["3.9"]
cash_rows = {
    "ni": ("National insurance contributions", "ni_employee"),
    "vat": ("Value added tax", "vat"),
    "fuel_duties": ("Fuel duties", "fuel_duty"),
    "capital_gains_tax": ("Capital gains tax", "capital_gains_tax"),
    "sdlt": ("Stamp duty land tax", "stamp_duty_land_tax"),
}

In the March 2026 EFO tables — the vintage currently configured in sources.yaml — sheet 3.8 is "Current receipts (on a cash basis)" and sheet 3.9 is "APD forecast - projection of passenger numbers". None of the five labels exist on 3.9, so every _find_row call raises ValueError, is caught, logs a warning, and appends no target.

The result is that all five cash-basis receipts targets are missing from the calibration surface — NICs, VAT, fuel duties, capital gains tax and SDLT — with nothing failing. Verified against the configured workbook:

Label	Sheet 3.8	Sheet 3.9
National insurance contributions	row 12	not found
Value added tax	row 13	not found
Fuel duties	row 21	not found
Capital gains tax	row 22	not found
Stamp duty land tax	row 24	not found

Note the existing cols_39/fy_39 column mapping (D→2024-25 … J→2030-31) already matches 3.8's layout exactly, which suggests the code was written against 3.8 and the sheet was renumbered by a later vintage without the reference being updated.

Fix

1. Locate the worksheet by title ("current receipts" in cell B2) rather than by number, so a future renumbering cannot silently repeat this.
2. Raise if none of the five rows are found. A per-row warning is right when one label changes; producing zero receipts targets is never a correct outcome and should not pass quietly.

Verification

Against the March 2026 receipts workbook the fix resolves sheet 3.8 and recovers all five rows (2026-27, £bn):

National insurance contributions   210.45
Value added tax                    187.70
Fuel duties                         24.63
Capital gains tax                   20.80
Stamp duty land tax                 18.15

Three unit tests cover the March 2026 layout, a hypothetical future renumbering, and the loud-failure path.

Why I noticed

Capital gains came up in a CGT reform analysis: the Enhanced FRS baseline produces ~1.29m CGT taxpayers holding £112bn of gains against HMRC's 378k and £66bn, and CGT revenue of ~£25bn against the OBR's £20.8bn forecast for 2026-27. Tracing why the OBR figure wasn't constraining the calibration led here. Restoring this target should pull CGT revenue toward the forecast directly.

Separately, populace has no capital gains targets at all in its UK surface — PolicyEngine/populace#459 — so gains totals and taxpayer counts remain unconstrained even with this fixed. The two are complementary: this PR restores the OBR revenue target; that issue is about HMRC gains and taxpayer targets.