Stream reform validation simulations

[MaxGhenis]

Summary

• avoid retaining every OBBBA stacked reform simulation during reform validation
• add a regression test that fails if intermediate simulations remain live across stacked states

Validation

• uv run ruff format packages/populace-build/src/populace/build/us_runtime/reform_validation.py packages/populace-build/tests/test_reform_validation.py
• uv run ruff check packages/populace-build/src/populace/build/us_runtime/reform_validation.py packages/populace-build/tests/test_reform_validation.py
• uv run pytest packages/populace-build/tests/test_reform_validation.py

This is needed before rerunning the full sparse 57k release validation: the previous full builder run completed calibration and wrote the H5/diagnostics, then was killed with exit code 137 during reform_validation.json generation after holding multiple large Microsimulation objects.