Pro-rate new_state_pension by reported amount

[MaxGhenis]

Addresses issue #1632.

Summary

The previous new_state_pension formula paid the flat max rate (eligible * p.amount * WEEKS_IN_YEAR) to every NEW-type retiree, regardless of actual NI record length or pre-2016 SERPS/S2P accrual.

This switches to the same share-of-max pattern that basic_state_pension already uses: share = reported / max_new_data_year, scaled by the current period's max.

Why it matters

• Partial NI records (< 35 qualifying years): retirees who in reality get pro-rated NSP were previously paid the full flat rate. The model now tracks their reported amount.
• Protected Payment: retirees with substantial pre-2016 SERPS/S2P accrual have a starting amount that exceeds the flat new rate; the excess is the Protected Payment component of NSP. The old formula clamped everyone at p.amount, silently dropping Protected Payment entirely. The new formula allows share > 1 so it flows through.

Together these close part of the ~£12 bn residual state-pension gap vs the OBR target tracked in #1632.

Tests

Five new YAML cases covering under-max / exact-max / above-max / non-NEW / zero-max guard. Full 976-case policy suite still passes.

Generated with Claude Code

[MaxGhenis]

Updated in response to a careful review. Two principled changes, one test addition, one comment fix:

1. NSP now caps share at 1 (min_(reported, max) / max) — mirrors basic_state_pension. The flat-rate component stays bounded by statute; the pre-2016 SERPS/S2P Protected Payment is captured as an add-on in additional_state_pension (now extended to NEW type). This keeps the variable names semantically clean: new_state_pension is always the headline flat-rate component, additional_state_pension is always the earnings-related top-up, regardless of type.

2. ASP extended to NEW-type retirees using the NSP max as the ceiling (was BASIC-only). Symmetric with BASIC's SERPS/S2P handling.

3. Added test for state_pension_reported = 0 on a NEW-type retiree — returns zero for both variables. This is the biggest behavioural change vs the old formula, which paid the flat max regardless; now deferral and FRS non-report both correctly yield zero.

4. Fixed stale 9627.80 comment to the correct 2021 max of 9339.20 (185.15 was the 2022 rate).

On the aggregate risk the reviewer flagged: rules should match statute, calibration should handle aggregate alignment. If the formula change shifts the NSP total vs OBR, that's something the weight-calibration pipeline can correct via upweighting NEW-type retirees; it shouldn't be patched by keeping a statutorily-wrong formula. Verifying actual aggregate impact remains useful but isn't a blocker.

Full 978-case policy suite still passes.