Align Medicare Part B premium modeling with enrollment targets

[daphnehanse11]

Summary

• Add gross_medicare_part_b_premium_if_enrolled for enrollment-consistent CMS premiums-from-enrollees calibration targets.
• Add medicare_irmaa_magi_two_years_prior as an explicit current-year input with fallback to AGI plus tax-exempt interest from two years prior.
• Route Part B and Part D IRMAA through the shared lagged MAGI variable and use the enrollment-gated gross Part B premium in Medicare cost.

Tests

• make format
• uv run pytest policyengine_us/tests/policy/baseline/gov/hhs/medicare/test_part_b_msp_offset.py
• uv run policyengine-core test policyengine_us/tests/policy/baseline/gov/hhs/medicare/eligibility/income_adjusted_part_d_premium_surcharge.yaml -c policyengine_us
• uv run pytest policyengine_us/tests/test_system_import.py

Closes #8677

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 100.00%. Comparing base (e59df75) to head (753003d).
⚠️ Report is 6 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff            @@
##              main     #8697   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files            1         5    +4     
  Lines           13        86   +73     
=========================================
+ Hits            13        86   +73     

Flag	Coverage Δ
unittests	100.00% <100.00%> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[MaxGhenis]

Reviewed the full diff against issue #8677 and checked the mechanics rather than trusting the summary (agent-assisted). This is clean.

What I verified:

• The medicare_irmaa_magi_two_years_prior extraction is exactly behavior-preserving: the old inline tax_unit(tax_exempt_interest_income, prior_period) already summed the person-level component to the tax unit, so the new add(tax_unit, prior_period, [...]) is identical. Both IRMAA consumers route through it, and a grep confirms no other file still inlines the two-years-prior pattern.
• gross_medicare_part_b_premium_if_enrolled isn't a bare wrapper — defined_for = medicare_enrolled is the point, gating the gross premium to enrolled beneficiaries so CMS premiums map to an enrollment-consistent calibration total.
• Test values hand-check out: MAGI 106,002 → 3,108 (a deliberate $1-over-threshold boundary), 500,000 → 7,546.80, Part D 500,000 → 1,029.60, and medicare_cost 12,280/10,060. Params are untouched here but line up with CMS 2026 (standard 202.90, first single IRMAA threshold above 109k).

One note: the medicare_cost switch to the enrollment-gated premium is numerically inert today (it's already defined_for = medicare_enrolled, and the two premiums are equal on that subset) — harmless, good for clarity.

I retyped the changelog fragment from .changed to .added on the branch since this adds two public variables (minor bump).

Heads up that #8532 also appends to test_part_b_msp_offset.py, so whichever lands second needs a trivial rebase; base is ~160 commits behind main, worth a rebase before merge.

Approving.