Add Disabled Students' Allowance

[MaxGhenis]

Summary

• add a first-pass Student Finance England Disabled Students' Allowance model
• use the published annual cap with explicit eligible-cost input rather than inventing a hidden award curve
• add a narrow default qualifying-condition proxy with explicit override, and wire DSA into program metadata and benefit aggregates

Policy shape

This is intentionally a partial model.

GOV.UK publishes the annual DSA cap and the main eligibility rules, but actual awards depend on an individual needs assessment rather than a public formula. So this PR models:

• England coverage via the existing Student Finance England proxy
• default course eligibility via higher-education evidence, with explicit override
• default qualifying-condition evidence via disability-benefit receipt or explicit blindness input, with explicit override for broader DSA-qualifying conditions
• explicit equivalent-support exclusion
• explicit annual eligible study-support costs, capped at the published annual maximum

It does not try to invent a synthetic average award curve.

Validation

Local checks passed:

• uvx ruff check policyengine_uk/variables/gov/dfe/disabled_students_allowance policyengine_uk/variables/gov/gov_spending.py policyengine_uk/variables/household/income/household_benefits.py policyengine_uk/variables/household/income/hbai_benefits.py
• uv run policyengine-core test policyengine_uk/tests/policy/baseline/gov/dfe/disabled_students_allowance/disabled_students_allowance.yaml

Focused policy tests cover:

• reimbursement below the cap
• cap at the published annual maximum
• equivalent-support exclusion
• the intentionally narrow default condition proxy
• explicit course and condition overrides

Current baseline

On the local enhanced_frs_2023_24 check at 2025:

• disabled_students_allowance spend: £0
• positive recipients: 0
• course-eligible share: 2.29%
• default qualifying-condition proxy share: 7.34%
• overlap share: 0.088%

That zero baseline is expected here because the amount side is explicit-cost driven and the survey does not observe DSA needs-assessment costs. So this PR adds the correct policy surface, but it is not calibration-ready on its own yet.