OpenConceptLab/ocl_online#89 | Support multiple AI Assistant outputs per input row

[snyaggarwal]

Linked Issue

Closes OpenConceptLab/ocl_online#89

[paynejd]

Reviewed with @paynejd. Four required changes before merge — inline suggestions cover the MapProject.jsx side; the display change is below.

1. Capture output_locale on every entry — it's part of the request payload and meaningfully affects the response, so it belongs on the saved record and in the attribute strip.
2. Drop the URI line from the visual display — once the resolved version is shown next to the template key, URI is redundant. Keep prompt_template_uri in the data model (audit log + JSON viewer still see it).
3. Read the resolved prompt version back from the invoke response (response.data.template.version). The intent is to record which version the AI Assistant used to generate the response, not which version (if any) we requested — clients can invoke with no version, let the assistant resolve, and read it back from the response.
4. Drop store-side dedup entirely — keep the full per-row history of AI responses. Move the "fire once" gate into the auto-match path only: auto-match fires once per row; a user can manually request as many additional AI recommendations as they want, even with the same (model, prompt, locale) combo.

---

Display side — AICandidatesAnalysis.jsx (outside the diff hunks, so written out here): replace the existing URI block (analysis?.prompt_template_uri && <span>…URI:…</span> around line 128) with an output_locale block:

{
  analysis?.output_locale &&
    <span style={{marginRight: '4px', display: 'inline-flex'}}>
      <Typography gutterBottom sx={{ color: 'text.secondary', fontSize: 12, mb: 0 }} component='span'>
        {t('map_project.output_locale')}:
      </Typography>
      <Typography gutterBottom sx={{ color: 'text.primary', fontSize: 12, mb: 0 }} component='span'>
        {analysis.output_locale}
      </Typography>
    </span>
}

Add a map_project.output_locale translation key to en/es/zh alongside it.

---

Backend verification: please confirm the new array-shaped analysis field round-trips cleanly through save→reload end-to-end. The load path defensively wraps singletons into [v], but I didn't verify the save path or the ocl-online DB column accept the array shape. A quick smoke test (create project → run AI analysis → save → reload → re-run on same row → confirm both entries persist and the pager shows 2/2) before merge would be reassuring.