feat: extend amount-from-flags to DLA and Attendance Allowance

[vahid-ahmadi]

Summary

Continues #44 by extending the PIP-from-flags pattern (PR #56) to:

• DLA care component — low / mid / high (SSCBA 1992 Sch.2 para.2)
• DLA mobility component — low / high (SSCBA 1992 Sch.2 para.3)
• Attendance Allowance — low / high (SSCBA 1992 s.64)

Synthetic households that set the eligibility flags now produce non-zero amounts; FRS-recorded amounts continue to pass through unchanged.

sim = Simulation.from_situation({
    "people":     {"c": {"age": 10, "dla_care_high": True, "dla_mob_high": True}},
    "benunits":   {"b": {"members": ["c"]}},
    "households": {"h": {"members": ["c"], "region": "London"}},
}, year=2025)
sim.run_microdata().benunits.loc[0, "baseline_total_benefits"]
# £9,747.40 of DLA + Child Benefit (when the benunit claims CB)

What's included

Engine (src/parameters/mod.rs, src/variables/benefits.rs):

• DlaParams struct (5 weekly rates) and AaParams struct (2 weekly rates)
• dla_care_amount(p, params), dla_mobility_amount(p, params), attendance_allowance_amount(p, params) helpers — same recorded-overrides-flag pattern as PIP
• passthrough_benefits rewired to use the helpers

Parameters (parameters/2025_26.yaml) — rates from 7 April 2025 per gov.uk:

benefit	low	mid	high
DLA care	£29.20	£73.90	£110.40
DLA mobility	£29.20	—	£77.05
Attendance Allowance	£73.90	—	£110.40

Python wrapper (models.py, __init__.py): new DlaParams and AaParams classes under Parameters.dla and Parameters.aa.

Rust unit tests (10): per-band rate / recorded-override / no-flag / no-params / benunit passthrough flow for DLA care, DLA mobility, and AA.

YAML policy-test cases (4) in tests/policy/dla_aa.yaml: child on DLA care+mobility high, child on DLA care low, AA high (post-SP-age, new SP cohort), AA low (pre-SP-age). All with would_claim_*: false to isolate the disability passthrough from modelled means-tested benefits.

Verified locally

• cargo test: 159 passing (149 + 10 new)
• pytest interfaces/python/tests: 83 passing
• python -m policyengine_uk_compiled.yaml_tests tests/policy: 25/25
• End-to-end: child with DLA care high + mob high produces £5,740.80 + £4,006.60 = £9,747.40 in passthrough_benefits, matching the rates exactly

Stacking

vahid/dla-aa-from-flags ← vahid/pip-from-flags (#56) ← vahid/lbtt-ltt (#55) ← vahid/yaml-test-harness (#54) ← vahid/parity-harness (#53) ← vahid/from-situation (#52). Six-deep stack.

Out of scope (remaining for #44)

• Reform-scaling of FRS-recorded amounts (mirroring the state-pension old_basic_pension_weekly baseline-scaling pattern)
• ADP (Adult Disability Payment, Scotland) and CDP (Child Disability Payment, Scotland) — fields exist on Person but no params
• PIP / DLA eligibility derivation from FRS disability indicators (currently flags are read directly from FRS recorded benefit amounts in src/data/clean.rs)
• LCWRA / disability premium passporting
• Carer premium expansion

Test plan

• CI passes (cargo + pytest + parity smoke + YAML cases)
• Manual sanity-check: setting dla_care_high: true on a synthetic household yields ~£5,741/yr DLA care
• FRS households unchanged

🤖 Generated with Claude Code

[vahid-ahmadi]

Rebased onto main — ready to merge.

Was the 6th branch in the stack; now independent on main and mergeable on its own. The DLA care/mobility and Attendance Allowance helpers mirror the established amount-from-flags pattern (recorded amount overrides flag, highest-rate fallthrough, weekly×52, graceful no-params handling), and the 2025/26 weekly rates match gov.uk for 7 April 2025. Fully additive (Option params + serde defaults; Person flag fields already exist), so FRS-recorded households are unchanged. Verified after rebase: builds clean, all 149 tests pass including the 11 DLA/AA tests, Python wrapper imports cleanly.

(During the rebase I kept this PR scoped to DLA/AA only — the PIP scaffolding it had been sharing with #56 stays in #56 — and converted one reform-flow test to its DLA equivalent so the coverage still holds on a standalone branch.)