[Epic] Prompt-acceptance data quality: distinguishable, joinable provenance (server-side)

Server-side half of the acceptance-data joinability effort (the UI / CJA half lives in adobe/project-elmo-ui#3108). Makes the prompts store's accept / dismiss / provenance data clean and joinable so acceptance can be measured on a dashboard.

## Issues (this repo)
- [ ] #3224 — Distinguish a real user-dismiss from a pipeline regeneration (`status=deleted` conflation)
- [ ] #3225 — Add an `entry_channel` dimension (onboarding-seed vs recommendation-accept vs manual vs CSV)

## Partner work in adobe/project-elmo-ui (epic #3108)
- Persist the Postgres `prompt_id` on the Helix card → native 3-way join (project-elmo-ui#3111)
- `decided_at` on accept/dismiss (project-elmo-ui#3109)

This is an epic rather than a single implementation task. Start with #3224 and #3225 for the server-side work, then review the linked project-elmo-ui epic and its related issues for the partner-side join requirements. Done means the acceptance data is distinguishable, includes entry-channel provenance, and supports the documented joins.

Written by the indexing model from the issue text.