Chain MicroImpute donor blocks

[MaxGhenis]

Summary

• update the PolicyEngine extra to published microimpute 3.1.0, which contains sequential ZeroInflatedImputer support
• fit one regime-aware ZeroInflatedImputer per compatible donor block instead of one independent wrapper per target
• group compatible donor targets into chained blocks while keeping the dividend composition block and entity policies intact
• add a regression proving the second target in a Microplex block uses the first target as a microimpute predictor

Why

The prior MP replacement build called microimpute one target at a time, so prior imputations were never available as predictors for later income/tax leaves. This PR makes that failure mode test-visible: multi-target regime-aware fits must share one fitted microimpute result, and target 2 must store target 1 in its predictor list.

Verification

• env -u UV_FROZEN uv run --python 3.13 --extra dev --extra policyengine ruff check src/microplex_us/pipelines/donor_imputers.py src/microplex_us/variables.py tests/pipelines/test_regime_aware_donor_imputer.py tests/test_variables.py
• env -u UV_FROZEN uv run --python 3.13 --extra dev --extra policyengine pytest -q tests/pipelines/test_regime_aware_donor_imputer.py tests/test_variables.py
• env -u UV_FROZEN uv run --python 3.13 --extra dev --extra policyengine pytest -q tests/test_pe_source_impute_specs.py tests/test_pe_source_impute_engine.py
• env -u UV_FROZEN uv run --python 3.13 --extra dev --extra policyengine pytest -q tests/pipelines/test_zi_qrf_backend.py
• env -u UV_FROZEN uv run --python 3.13 --extra dev --extra policyengine pytest -q tests/pipelines/test_donor_imputer_negative_preservation.py tests/data_sources/test_scf_net_worth_components.py -k "imputer or regime_aware or columnwise"
• installed-version smoke confirms microimpute 3.1.0 and ZeroInflatedImputer(..., sequential=True)