Add Stage 1 checkpoint reuse boundary

[anth-volk]

Fixes #1074

Summary

• Add checkpoint store, persisted reuse manifest, and rerun planner abstractions for Stage 1 dataset build substeps.
• Require semantic identity manifest matches before physical checkpoint files can be reused, while preserving the existing checkpoint file layout.
• Key persisted reuse identities by execution-unit identity_key while keeping substep_id as the public status grouping, so multiple command/script units inside one Stage 1 substep do not overwrite each other.
• Record checkpoint/reuse decisions and reuse reasoning in substep results, status events, and output contract metadata.
• Add Stage 1 AI-facing docs for reuse identity granularity, conditional running, checkpoint metadata, and documentation-sync expectations.

Validation

• ruff check policyengine_us_data/build_datasets/rerun.py modal_app/data_build.py tests/unit/test_build_dataset_rerun.py tests/unit/test_build_dataset_checkpoints.py tests/unit/test_modal_data_build.py
• ruff format --check policyengine_us_data/build_datasets/rerun.py modal_app/data_build.py tests/unit/test_build_dataset_rerun.py tests/unit/test_build_dataset_checkpoints.py tests/unit/test_modal_data_build.py
• uv run --no-sync pytest tests/unit/test_build_dataset_checkpoints.py tests/unit/test_build_dataset_rerun.py tests/unit/test_dataset_build_stage_contract.py tests/unit/test_modal_data_build.py tests/unit/test_pipeline_doc_guards.py tests/unit/test_pipeline_docs_extractor.py
• uv run --no-sync --with pyyaml python scripts/run_quality_guards.py
• uv run --no-sync --with pyyaml --with pytest pytest tests/unit/test_pipeline_docs_extractor.py tests/unit/test_pipeline_doc_guards.py