CI workflow efficiency improvements

[juaristi22]

Fix #477
Fix #470

Changes (data_build.py)

1. Parallel Execution. Data builds now run in dependency-based phases:

• Phase 1 (parallel): uprating.py, acs.py, irs_puf.py
• Phase 2 (parallel): cps.py, puf.py
• Phase 3: extended_cps.py
• Phase 4 (parallel): enhanced_cps.py, create_stratified_cps.py
• Phase 5: small_enhanced_cps.py

2. Checkpointing System (more robust to preemptions)

• Creates data-build-checkpoints Modal Volume for persistent storage
• Per-branch checkpoint paths (/checkpoints/{branch}/) prevent cross-PR contamination
• After preemption + restart, completed scripts are skipped (restored from checkpoint)
• Tests run module-by-module with progress tracking for mid-test-suite resume
• Checkpoints cleaned up after successful completion
• New --clear-checkpoints flag to force fresh builds

3. Duplicate Test Removal

• Previously, local area calibration tests ran twice (once explicitly, once as part of full suite)
• Now each test module runs exactly once

Additions (create_representative_fixture.py)

1. Smaller, Comprehensive District-Level Dataset

• Creates a ~4k household dataset with all CDs and most counties represented.
• Can be swapped into conftest.py for faster build and calibration testing