[codex] Use modeled Medicare Part B inputs and targets

[MaxGhenis]

What changed

• store CPS PEMCPREM as medicare_part_b_premiums_reported and carry medicare_enrolled explicitly into the built datasets
• stop QRF-imputing Medicare Part B premiums on the clone half
• stop allocating generic PUF medical expenses into Part B now that Part B is modeled in policyengine-us
• document the national Part B calibration target as an approximate beneficiary-paid out-of-pocket benchmark instead of gross CMS premium income
• add CMS target helper coverage plus input regression tests

Why

This is the data-side half of the modeled Part B baseline change. Once policyengine-us models Part B premiums directly, policyengine-us-data should feed the model the reported audit series and enrollment inputs, not overwrite Part B with copied or imputed values. The target metadata also needs to reflect out-of-pocket semantics rather than gross trust-fund premium income.

Impact

• future CPS/extended/enhanced builds use modeled Part B instead of copied/imputed Part B
• clone-half medical-expense inputs are less conceptually inconsistent
• calibration metadata is clearer about what medicare_part_b_premiums is trying to match

Validation

• uv run pytest -q tests/unit/test_medicare_part_b_inputs.py tests/unit/test_cms_medicare_targets.py
• uv run towncrier check --compare-with upstream/main
• git diff --check

Paired model PR: PolicyEngine/policyengine-us#7971

[vercel]

@MaxGhenis is attempting to deploy a commit to the PolicyEngine Team on Vercel.

A member of the Team first needs to authorize it.

[MaxGhenis]

Superseded by clean replacement PR #714 from codex/medicare-partb-data-clean.