fix(review): persist AI review diagnostics into cache metadata on degraded runs

[JSONbored]

Problem

ai_review_cache.metadata_json carries inconclusive — it records that a review degraded, but never why.

The per-attempt AiReviewDiagnostic values that hold the answer (missing_assessment, unparseable_output, empty_output, provider_error) were only ever passed to capturePostHogReviewFailure as a capture property. They reach PostHog and nowhere else — not the database, not the container logs.

That is a real dead end during an incident. The 2026-07-27 investigation into intermittently missing AI review summaries could not tell the failure classes apart: a docker logs grep over the incident window returns zero matches, and reaching PostHog needs an interactive OAuth an incident responder may not be able to complete.

Change

Persist the compact model#attempt:status[:error] strings into metadata_json alongside inconclusive, so the three classes are separable at read time — they are three different bugs with three different fixes:

• missing_assessment — the model genuinely returned no narrative
• unparseable_output / empty_output — it returned something parseModelReview could not read (no balanced JSON object)
• provider_error — the call itself failed

Scope and cost

Written on the degraded paths only, never on the healthy one. Every byte lands in a D1 row, and this database hit its 10 GB ceiling on 2026-07-26 — so this buys diagnosability for the small minority of reviews that actually failed rather than growing all of them.

No schema change: metadata_json is an existing column, so no migration and no generated-artifact regeneration.

Safety

The strings come from formatReviewDiagnosticsForCapture, whose error field is errorMessage() output and never raw provider text — the public/private boundary is unchanged. This metadata is read-only structured output, not a cache-replay trigger.

Tests

Both sides of the new branch are covered:

• degraded run persists diagnostics, and an empty provider response is recoverable from the row as empty_output
• healthy run omits them entirely (the D1-cost guarantee)

Patch lines are fully covered; the changed file's remaining uncovered lines (375, 597, 707, 875) all predate this change.

Refs #9432

[loopover-orb]

Important

🟨🟨🟨🟨🟨🟨🟨🟨🟨🟨🟨🟨

⏳ LoopOver is waiting…

LoopOver has seen this pull request and is waiting on CI checks to finish before reviewing it. This comment will update once the review runs.

_{🟩 Safe / merged · 🟦 Advisory · 🟨 Held for review · 🟥 Blocked / closed · 🟨 Waiting}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 88.64%. Comparing base (fb8e2eb) to head (9c3672c).
⚠️ Report is 8 commits behind head on main.
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #9447      +/-   ##
==========================================
- Coverage   89.54%   88.64%   -0.91%     
==========================================
  Files         843      843              
  Lines      110011   110012       +1     
  Branches    26184    26185       +1     
==========================================
- Hits        98511    97521     -990     
- Misses      10238    11520    +1282     
+ Partials     1262      971     -291     

Flag	Coverage Δ
backend	93.61% <100.00%> (-1.65%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
src/queue/ai-review-orchestration.ts	98.78% <100.00%> (+<0.01%)	⬆️

... and 3 files with indirect coverage changes

[superagent-security]

Superagent didn't find any vulnerabilities or security issues in this PR.