feat: compute PIP daily-living and mobility amounts from flags

[vahid-ahmadi]

Summary

Addresses part of #44. PIP amount fields are now computed from the existing eligibility flags + a new PipParams parameter struct. Stacked on #55.

Before: setting pip_dl_enh: true on a synthetic household produced £0 PIP because pip_daily_living was only ever populated from FRS recorded amounts.

After:

from policyengine_uk_compiled import Simulation
sim = Simulation.from_situation({
    "people":     {"p": {"age": 35, "pip_dl_enh": True, "pip_mob_std": True}},
    "benunits":   {"b": {"members": ["p"]}},
    "households": {"h": {"members": ["p"], "region": "London"}},
}, year=2025)
sim.run_microdata().benunits.loc[0, "baseline_total_benefits"]
# £7,259.20 (= £5,740.80 DL enhanced + £1,518.40 mobility standard,
#            plus modelled means-tested benefits as applicable)

Reforms flow through too: doubling daily_living_enhanced_weekly doubles the synthetic household's PIP DL amount.

What's included

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

• PipParams struct with the four 2025/26 weekly rates
• pip_daily_living_amount(p, params) and pip_mobility_amount(p, params) helpers:
  • Recorded amount > 0 → pass through unchanged (preserves FRS calibration; an FRS-claimant on partial-year PIP keeps their reported amount)
  • Amount 0 + flag set → compute from params.pip × 52
  • No params loaded or no flag → return 0
• passthrough_benefits now uses the helpers, so PIP from flags flows into total_benefits and downstream net income

Parameters (parameters/2025_26.yaml): rates from 7 April 2025 per gov.uk/pip/what-youll-get under WRA 2012 s.79 / SI 2013/377:

component	std	enh
daily living	£73.90/wk	£110.40/wk
mobility	£29.20/wk	£77.05/wk

Python wrapper (models.py, __init__.py): new PipParams class exposed under Parameters.pip.

Rust unit tests (8): enhanced/standard/recorded-override/no-flag/no-params and reform-scaling for both DL and mobility, plus a passthrough-flow test on a benunit.

YAML policy-test cases (4) in tests/policy/pip.yaml: DL+mob combinations end-to-end through the binary, with would_claim_*: false on the benunit to isolate PIP from modelled means-tested benefits.

Verified locally

• cargo test: 149 passing (141 + 8 new)
• pytest interfaces/python/tests: 79 passing
• python -m policyengine_uk_compiled.yaml_tests tests/policy: 21/21
• End-to-end: synthetic household with pip_dl_enh + pip_mob_std produces £7,259.20 passthrough benefit (= £5,740.80 + £1,518.40, matching the 2025/26 weekly rates exactly)

Stacking

vahid/pip-from-flags ← vahid/lbtt-ltt (#55) ← vahid/yaml-test-harness (#54) ← vahid/parity-harness (#53) ← vahid/from-situation (#52). Five-deep stack; once each merges in order, this rebases onto main.

Out of scope (follow-ups for #44)

• Same flag-based computation for DLA, AA, and Carer's Allowance (each has its own flag set on Person)
• Reform scaling of FRS-recorded amounts (mirroring the state-pension old_basic_pension_weekly baseline-scaling pattern)
• PIP eligibility derivation from FRS disability indicators (currently the flags are read directly from FRS recorded benefit amounts in src/data/clean.rs)
• LCWRA / disability premium passporting beyond what already exists

Test plan

• CI passes (cargo + pytest + parity smoke + YAML cases)
• Manual sanity-check: setting pip_dl_enh: true on a synthetic household yields ~£5,741 baseline_total_benefits in addition to any modelled benefits
• FRS households unchanged (recorded amounts pass through, no scaling)

🤖 Generated with Claude Code

[vahid-ahmadi]

Rebased onto main — ready to merge.

Previously stacked behind the LBTT/PIP chain; now sits directly on main and is mergeable independently. The PIP daily-living/mobility helpers are correct: enhanced-before-standard precedence, recorded-amount passthrough (FRS values preserved), graceful no-params handling, weekly×52. The four 2025/26 weekly rates match gov.uk exactly. Fully additive — Option param with serde(default), so existing param files are unaffected. Verified: Rust builds, PIP unit tests pass, Python wrapper imports cleanly. Test coverage is strong (8 unit tests + YAML cases).