Continue an interrupted Sentinel run.

If the user gave no run name, list .sentinel/runs/*/state.json. A run without an outcome field is still open. Show each open run's slug, tier and last completed stage, and ask which to continue.
Run sentinel resume <slug>. It prints the next stage, the pending stages, the rounds used, and any loop that hit its cap.
If the run has an outcome, it is finished; say so and stop.
If a loop is capped, do not continue. Say the increment is not production-ready, list the open findings from the latest result files, and ask the user how to proceed.
Otherwise read .sentinel/runs/<slug>/decisions.md and the newest *-result.md files in that folder, then continue from the next stage by following the sentinel skill (its "Running a stage" and "Pipeline" sections). Do not redo stages already in completed, and re-verify the last stage's result before building on it.

sentinel-resume