Align triple lock with published DWP / OBR uprating rates (#953)

[vahid-ahmadi]

Summary

• The triple lock formula previously used OBR calendar-year average earnings, but DWP applies May-July AWE growth. That left every uprating year 0.5-1.2pp off vs published rates.
• Add outturn.yaml listing the published triple lock rates for 2022-2026 (3.1%, 10.1%, 8.5%, 4.1%, 4.8%) and update create_triple_lock.py to prefer outturn over the max-of formula when available.
• Correct the hard-coded April 2025 basic / new state pension amounts (£172.38 / £224.96 → £176.45 / £230.25) and add the April 2026 figures (£184.90 / £241.30) so basic and new state pension amounts now match DWP / OBR exactly through 2026-27. The formula still drives 2027+.

Closes #953.

Test plan

• python -m pytest policyengine_uk/tests/test_triple_lock_outturn.py -v — 11 new tests pass (5 triple lock yoy, 3 basic SP weekly, 3 new SP weekly).
• make format — clean.
• Reviewer to spot-check uprating chain through gov.economic_assumptions.indices.triple_lock for 2027+ years.

🤖 Generated with Claude Code

[MaxGhenis]

Verified all six state-pension weekly amounts and both uprating rates against gov.uk benefit-and-pension-rates publications and the April 2026 DWP announcement. 176.45, 184.90, 230.25, 241.30 all match published outturn; 0.041 (2025-26) and 0.048 (2026-27) round cleanly to the published amounts.

Minor inaccuracy in the comment that doesn't block merge: the outturn.yaml description and the code comment imply the triple lock uprating follows a single methodology, but DWP's choice between AWE total-pay and regular-pay varies year to year — 4.1% for 2025-26 matches AWE total pay per the Sept 2024 DWP announcement, while 4.8% for 2026-27 matches AWE regular pay (excl. bonuses) per the 2025 uprating. Worth a follow-up to soften the methodology claim to 'AWE (measure DWP selects each year)'. Numbers are right either way.

Approach (outturn override + formula fallback for forecast years) is clean and exactly right.