Add PUF support clone for eCPS replacement

[MaxGhenis]

Summary

• add an eCPS-style PUF support clone stage that runs before later donor imputations
• enable the stage by default for the PolicyEngine US data rebuild path, with a CLI override
• add calibration activation diagnostics and source-weight diagnostics for zero-initial-weight PUF clone households
• cover generated entity IDs, no-calibration runs, donor ordering, and configured clone flags in regression tests

Closes #140.

Validation

• uv run --python 3.13 ruff check src/microplex_us/pipelines/us.py src/microplex_us/pipelines/artifacts.py src/microplex_us/pipelines/pe_us_data_rebuild.py src/microplex_us/pipelines/pe_us_data_rebuild_checkpoint.py tests/pipelines/test_us.py tests/pipelines/test_artifacts.py tests/pipelines/test_pe_us_data_rebuild.py tests/pipelines/test_pe_us_data_rebuild_checkpoint.py
• uv run --python 3.13 ruff format --check src/microplex_us/pipelines/us.py src/microplex_us/pipelines/artifacts.py src/microplex_us/pipelines/pe_us_data_rebuild.py src/microplex_us/pipelines/pe_us_data_rebuild_checkpoint.py tests/pipelines/test_us.py tests/pipelines/test_artifacts.py tests/pipelines/test_pe_us_data_rebuild.py tests/pipelines/test_pe_us_data_rebuild_checkpoint.py
• uv run --python 3.13 --extra dev python -m pytest tests/pipelines/test_artifacts.py::TestSaveUSMicroplexArtifacts::test_source_weight_diagnostics_respects_configured_puf_flag_column -q
• uv run --python 3.13 --extra dev --extra policyengine python -m pytest tests/pipelines/test_us.py -k 'puf_support_clone or integrate_donor_sources' -q\n- uv run --python 3.13 --extra dev --extra policyengine python -m pytest tests/pipelines/test_artifacts.py tests/pipelines/test_pe_us_data_rebuild.py tests/pipelines/test_pe_us_data_rebuild_checkpoint.py -q\n\n## Review\n- /cycle completed; final independent reviewer reported no actionable findings.

[MaxGhenis]

PUF aggregate-record follow-up added after comparing to eCPS PR #627.

What changed:

• MP now disaggregates PUF aggregate rows (MARS=0, RECID 999996-999999) before PUF source construction instead of dropping them.
• Local raw PUF smoke: 207,696 raw rows with 4 aggregate rows -> 207,814 loaded rows, 122 synthetic aggregate-derived rows, and no remaining MARS=0 rows.
• This mirrors the relevant eCPS upstream behavior from Disaggregate PUF aggregate records and fix QRF high-income training policyengine-us-data#627 while keeping MP support-clone architecture separate.

Additional validation:

• uv run --python 3.13 ruff check src/microplex_us/data_sources/puf.py src/microplex_us/pipelines/artifacts.py tests/test_puf_source_provider.py
• uv run --python 3.13 ruff format --check src/microplex_us/data_sources/puf.py src/microplex_us/pipelines/artifacts.py tests/test_puf_source_provider.py
• uv run --python 3.13 --extra dev python -m pytest tests/test_puf_source_provider.py::test_load_puf_raw_disaggregates_aggregate_records tests/pipelines/test_artifacts.py::TestSaveUSMicroplexArtifacts::test_source_weight_diagnostics_respects_configured_puf_flag_column -q

A disaggregated Gate-1 artifact rebuild is running locally now; the earlier no-disagg support-clone comparison is diagnostic only.