Filter invalid SIPP imputation weights

[MaxGhenis]

Summary

• filter non-positive/non-finite SIPP donor weights before QRF fits with microimpute target filters
• apply the guard to legacy SIPP tip/asset/vehicle models and calibration source-impute SIPP models
• add regression coverage for target-filter reindexing and the SIPP tip failure seen in Modal

Failure addressed

Latest publication after #1103 failed in Modal Stage 1:

• GitHub candidate run: 26255866421
• GitHub pipeline run: 26255880509
• Modal run ID: usdata-gha26255866421-a1
• Modal app: us-data-1-115-5-patch-usdata-gha26255866421-a1
• Source SHA: e6fdcaf
• Error: policyengine_us_data/datasets/cps/cps.py -> train_tip_model() -> microimpute ValueError: Weights must be positive and finite; found 8 non-positive or NaN weight(s)

Tests

• uv run ruff format policyengine_us_data/utils/source_quality.py policyengine_us_data/datasets/sipp/sipp.py policyengine_us_data/calibration/source_impute.py tests/unit/test_source_quality.py tests/unit/datasets/test_sipp_tip_columns.py
• uv run ruff check policyengine_us_data/utils/source_quality.py policyengine_us_data/datasets/sipp/sipp.py policyengine_us_data/calibration/source_impute.py tests/unit/test_source_quality.py tests/unit/datasets/test_sipp_tip_columns.py
• uv run pytest tests/unit/test_source_quality.py tests/unit/datasets/test_sipp_tip_columns.py tests/unit/calibration/test_source_impute.py -q
• uv run python - <<'PY'
  from policyengine_us_data.datasets.sipp.sipp import train_tip_model
  model = train_tip_model()
  print(type(model).name)
  PY