Add canonical US runtime stage contracts and docs

[anth-volk]

Summary

Implements the canonical 9-stage US Microplex runtime taxonomy proposed in #20.

Closes #42.

This adds:

• microplex_us.pipelines.stage_contracts for stable stage definitions, artifact contracts, validation placeholders, diagnostics, and resume posture.
• microplex_us.pipelines.stage_manifest for saved-run stage manifests, validation evidence manifests, and Stage 6 entity checkpoint helpers.
• artifact saving updates that write stage_manifest.json, source-plan, calibration-summary, PE entity checkpoint metadata, and validation evidence sidecars.
• data-flow snapshot and PE-US-data checkpoint evidence updates that use the canonical Stage 9 validation/benchmarking stage.
• Jupyter Book/Sphinx docs for the stage taxonomy, stage contracts/manifests, and API reference.

Out of scope

• Automatic conditional execution or a scheduler.
• A shared stage-validation runtime.
• Full source-frame snapshotting for every raw input source.
• Changes to modeling, synthesis, or calibration behavior beyond saved metadata/artifacts.

Validation

• ruff check src/microplex_us/pipelines/stage_contracts.py src/microplex_us/pipelines/stage_manifest.py src/microplex_us/pipelines/artifacts.py src/microplex_us/pipelines/data_flow_snapshot.py src/microplex_us/pipelines/pe_us_data_rebuild_checkpoint.py src/microplex_us/pipelines/__init__.py tests/pipelines/test_stage_contracts.py tests/pipelines/test_stage_manifest.py tests/pipelines/test_data_flow_snapshot.py tests/pipelines/test_artifacts.py tests/pipelines/test_pe_us_data_rebuild_checkpoint.py tests/pipelines/test_backfill_pe_native_audit.py
• PYTHONPATH=/private/tmp/microplex-comparison/microplex/src:/private/tmp/microplex-comparison/microplex-us/src python -m py_compile src/microplex_us/pipelines/stage_contracts.py src/microplex_us/pipelines/stage_manifest.py src/microplex_us/pipelines/artifacts.py src/microplex_us/pipelines/data_flow_snapshot.py src/microplex_us/pipelines/pe_us_data_rebuild_checkpoint.py src/microplex_us/pipelines/__init__.py
• PYTHONPATH=/private/tmp/microplex-comparison/microplex/src:/private/tmp/microplex-comparison/microplex-us/src python -m pytest tests/pipelines/test_stage_contracts.py tests/pipelines/test_stage_manifest.py tests/pipelines/test_data_flow_snapshot.py (9 passed, 1 skipped; skipped parquet roundtrip because local env lacks pyarrow)
• PYTHONPATH=/private/tmp/microplex-comparison/microplex/src:/private/tmp/microplex-comparison/microplex-us/src python -m pytest tests/pipelines/test_pe_us_data_rebuild_checkpoint.py::test_attach_policyengine_us_data_rebuild_checkpoint_evidence_updates_manifest tests/pipelines/test_pe_us_data_rebuild_checkpoint.py::test_attach_policyengine_us_data_rebuild_checkpoint_evidence_registers_calibration_only_runs tests/pipelines/test_backfill_pe_native_audit.py (6 passed)
• PYTHONPATH=/private/tmp/microplex-comparison/microplex/src:/private/tmp/microplex-comparison/microplex-us/src jupyter-book build docs (passes; existing docs emit 206 warnings from pre-existing untoc'd pages/local absolute links)

Local limitation

The existing artifact save path still cannot run end-to-end in this local environment because pandas parquet writes require pyarrow or fastparquet, neither of which is installed here.

[MaxGhenis]

I opened #45 as a small stacked follow-up for the run-contract/stage-manifest bridge.

Resolution there: keep this PR's 9 canonical stage IDs as the saved-run taxonomy, and map older operational run-contract IDs into that taxonomy when the dashboard reads run_summary.json. The dashboard preserves legacy labels as provenance via legacy_stage_id / legacy_completed_stages, so existing run summaries keep their debug value without creating a parallel lifecycle.