Cover exported income inputs in PUF top-tail guard

[MaxGhenis]

Summary

• extend the PUF support-clone top-tail rough AGI proxy to count exported/formula input aliases that appear in the final H5, including employment_income_before_lsr, self_employment_income_before_lsr, disaggregated pension fields, and Social Security components
• keep scaling limited to PUF-imputed investment/business components, so CPS wage/support fields remain fixed while the top-tail component is reduced enough to keep rough AGI below the open-ended SOI bin
• add a regression test matching the failed post-Prefer capital gains components in PUF top-tail guard #168 H5 shape: exported LTCG alias plus employment_income_before_lsr and pension/Social Security inputs

Evidence

The post-#168 rebuild still missed state/MD/adjusted_gross_income/amount/1000000_inf:

• MP latest-local loss: 1.2587150824491427
• eCPS latest-local loss: 0.7685371816961414
• largest target delta: state/MD/adjusted_gross_income/amount/1000000_inf, candidate estimate 0.6467862826845676 vs target 0.01419332389429203

A row-level check of the final H5 found the remaining Maryland top-tail tax unit at rough AGI about $79.069M: exported LTCG alias around $78.858M plus employment_income_before_lsr around $178k. The guard had capped the capital-gains component, but its rough AGI proxy did not count the exported LSR/base income fields used in the final dataset shape.

Validation

• uv run ruff check src/microplex_us/pipelines/us.py tests/pipelines/test_us.py
• uv run ruff format --check src/microplex_us/pipelines/us.py tests/pipelines/test_us.py
• uv run --extra dev python -m pytest tests/pipelines/test_us.py -k 'puf_support_clone_top_tail_guard'
• uv run --python 3.13 --extra dev --extra policyengine python -m pytest tests/pipelines/test_us.py (180 passed, expected warnings only)