Add compact PolicyEngine dataset export

[MaxGhenis]

Summary

• add a supported CLI to compact PolicyEngine H5 datasets by keeping the largest household weights
• preserve linked person/tax-unit/SPM/family/marital-unit rows and optionally rescale household weights to the source total
• cover entity filtering, external weight selection, CLI summary output, and mismatched-weight validation

Validation

• uv run --python 3.13 --extra dev ruff check pyproject.toml src/microplex_us/pipelines/compact_policyengine_dataset.py tests/pipelines/test_compact_policyengine_dataset.py
• uv run --python 3.13 --extra dev ruff format --check pyproject.toml src/microplex_us/pipelines/compact_policyengine_dataset.py tests/pipelines/test_compact_policyengine_dataset.py
• uv run --python 3.13 --extra dev --extra policyengine pytest -q tests/pipelines/test_compact_policyengine_dataset.py
• real-data smoke: compacted current MP candidate to 1,000 households and wrote summary under artifacts/mp300k_compact_screen_20260528

Notes

This supports the compact mp-size release path. Current screening found top-100k/top-120k compacts can meet size/runtime gates on the stored May 17 target matrix, but latest-us-data scoring still loses to eCPS because the current broad target surface is dominated by national IRS other and state AGI distribution regressions.