TRR BACKEND: Stage runner command (optional)

[therealityreport]

Goal: Allow operators to rerun a single stage without touching others.

Motivation: The orchestrator is sequential; without a stage runner, operators keep using --from/--to and risk unintended side effects.

TODO

• Add CLI command: python -m trr_backend.cli pipeline stage <stage_name> --run-id <id>
• Validate that run exists and stage is in valid state for rerun
• Handle manifest requirements (Stage 1 vs later stages)
• Update pipeline.run_stages row on completion/failure
• Add --dry-run support consistent with main orchestrator

Acceptance

• Operator can safely rerun a single failed stage without re-executing prior stages
• Stage runner respects same manifest/skip-s3 semantics as orchestrator