Add age-based bus fare allocation and young-person fare reform

[vahid-ahmadi]

What

Adds the person/age allocation layer that lets PolicyEngine model age-targeted bus fare policies (e.g. Scotland-style free travel for under-22s), on top of the household bus_fare_spending input.

• gov.dft.bus.fare_allocation_weight_by_age — provisional per-age weight for splitting household bus fares across members. NTS (England) bus-trip-by-age profile, adjusted for concessionary (free-pass) travel so it tracks fares paid, not trips. Clearly documented as a modelling allocation, not a fare-incidence estimate.
• gov.dft.bus.young_person_fare.{age_limit, rate} — policy levers. Inert by default (age_limit = 0), so baseline is unchanged. A reform sets age_limit = 22 for free under-22 travel; rate (default 0 = free) supports discounts/flat fares too.
• household_bus_fare_weight / person_bus_fare_spending — the allocation (household_fare × age_weight / household_total_weight), conserving the household total.
• bus_fare_relief — government-funded fares for eligible young people. Routed through dft_subsidy_spending, which already feeds both household_benefits (→ net income) and gov_spending (→ Exchequer cost), so both sides of the policy are captured with no core surgery.

Why

bus_fare_spending is household-level (LCFS has no person-level fares), and rail is modelled at household level too. An age-targeted reform needs the fare attributed to people, which this provides — following existing PE-UK household→person allocation patterns (housing_benefit_assessable_capital, employer_ni_response_consumer_incidence).

Tests

test_bus_fare_age_allocation.py (3 passing): allocation splits by age weight and conserves the total; baseline relief is zero; an under-22 reform relieves the eligible member's allocated fare.

Caveats / scope

• The age weights are provisional (two verified NTS anchors — 17–20 peak, ~25 mean — plus modelled mid-bands and assumed concessionary paid-shares; England used as a UK proxy). Fine for a directional prototype; publication-grade costing should use historical NTS0601 ODS or NTS microdata.
• Static — no behavioural response (lower fares → more trips) yet.

Stacking

Stacked on #1780 (add-bus-fare-spending-variable) — it references bus_fare_spending. Merge #1780 first; this PR's base will retarget to main automatically. Also relies on the imputation in policyengine-uk-data#428.

🤖 Generated with Claude Code

[MaxGhenis]

The modelling is sound and the reform is correctly inert by default (agent-assisted review): age_limit defaults to 0, so bus_fare_relief is zero for everyone and dft_subsidy_spending is unchanged at baseline. The household-to-person allocation conserves the household total, and routing relief through dft_subsidy_spending captures both the in-kind household benefit and the Exchequer cost without touching core. No double-count: bus_fare_spending has no other consumer on main, and bus fares are VAT zero-rated.

Two things before this can merge:

1. CI has never run on this branch (zero check runs), and it's 72 commits behind main. #1780 and uk-data#428 have both merged since, so rebase onto main and push to trigger CI — the diff still shows #1780's files because the branch predates that squash-merge, and those drop out on rebase. Not mergeable until CI is green.

2. The weights in fare_allocation_weight_by_age are provisional modelling assumptions, not measured values — only the 17-20 peak is anchored to NTS; the mid-bands and concessionary paid-shares are assumed, with England as a UK proxy. Fine for a directional prototype and documented as such, and it only affects the age-incidence of a reform (the household total is conserved). Flagging so we don't reuse these for a published costing without NTS0601 / NTS microdata.

Minor: test_bus_fare_spending_uses_cpi_uprating is sitting in test_road_fuel_volume_uprating.py — give it its own file. And consider whether young_person_fare belongs under gov.contrib rather than the baseline gov.dft.bus tree, since it models a non-UK-wide scheme that only switches on via reform; either works, but contrib signals "contributed lever" more clearly.

No blocking code changes — I'll approve once it's rebased and CI is green.