Create H5 files for local area microsimulations from calibrated weights

[baogorek]

Overview

After local area calibration produces weight vectors, we need to generate H5 dataset files that can be used for microsimulations at the congressional district level.

Requirements

• Create stacked H5 datasets from calibrated weights
• Support generating datasets for:
  • All 436 CDs in a single file
  • Per-state files (e.g., NY.h5 with all NY districts)
  • Per-CD files (e.g., NY-10.h5)
• Properly handle county assignment so that county-dependent variables (like in_nyc) work correctly
• Reindex entity IDs to prevent overflow when stacking many districts

Technical Details

• Input: Calibrated weight vector w of length n_households × n_cds
• Output: H5 files compatible with PolicyEngine microsimulation
• County assignment should use population-weighted P(county|CD) distributions from Census block data
• State variables must be updated and caches cleared for correct variable recalculation

Acceptance Criteria

• Stacked dataset builder creates valid H5 files
• County-dependent variables (e.g., in_nyc) return correct values
• Can generate per-state and per-CD subsets
• Unit tests pass