Rerun Outputs

Re-run all configured outputs for an interview (admin only).

Motivating use case: a setup-assistant onboarding webhook crashed mid- pipeline (e.g. a code bug) and the template was never created. After deploying the fix we want the outputs to retry without forcing the user to redo the conversation. Re-running outputs blindly does that — the setup-assistant webhook fires again and creates the template from the existing transcript + frames.

All outputs are re-run, including ones that previously succeeded. For webhook outputs this means the receiver may be hit twice. Admins are expected to know this and only invoke this when re-delivery is acceptable (or when the prior attempt clearly didn't side-effect, as is the case for an HTTP 500 from our own internal webhook handler).

Mirrors the dispatcher in /v1/sessions/{id}/result that runs at natural session end — the same _process_interview_background helper, same context-load logic, same persistence.