feat(review): detect successor-superseded reversals for the one-shot close culture

[JSONbored]

What

Adds the successor-based reversal detector (#8166): when a PR merges, scan the last 30 days of real (non-dry-run) bot closes in the same repo and record a reversal_superseded event — plus the per-rule reversed human-override events the calibration corpus consumes — for any bot-closed PR the merge supersedes.

Why

The gate's own one-shot design tells a wronged contributor "recovery = open a fresh PR", so the reopen-shaped reversal signal is structurally near-impossible here — verified against the full production ledger: zero reversal events ever recorded. This is the culture's actual "the bot was wrong" shape, and it is what finally feeds the corpus's positive class organically.

How

• src/review/reversal-superseded.ts — pure, conservative matcher: a shared linked issue (the strongest intent signal), or same author reworking ≥50% of the closed PR's files. Borderline records nothing; unknown authors and file-less closed PRs never match. Every recorded event carries the matched heuristics.
• recordSupersededReversals in outcomes-wire.ts — runs in the merged branch of the existing reversal wire; idempotent per closed target; every write best-effort; reuses the calibration: wire recordRuleFired/recordHumanOverride for the remaining isConfiguredGateBlocker codes (excludes linked_issue_scope_mismatch, covered by #8101) #8104/calibration: wire linked_issue_scope_mismatch into the shared signal-tracking module (fired + reversal) #8101 override recorders so fired rules against the superseded PR get their reversed verdict.
• Public reversal stats and the accuracy trend count the new event type alongside reversal_reopened/reversal_reverted.

Validation

• Full npm run test:ci green locally (1118 test files).
• 100% line+branch coverage on both touched src ranges, including the fail-safe arms (forced write rejections, undefined D1 result set, garbage target keys, self-close skip).

Closes #8166

[superagent-security]

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

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 89.63%. Comparing base (d89698d) to head (67ac8dc).
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #8188      +/-   ##
==========================================
- Coverage   92.07%   89.63%   -2.44%     
==========================================
  Files         769      101     -668     
  Lines       77882    23021   -54861     
  Branches    23529     3965   -19564     
==========================================
- Hits        71707    20635   -51072     
+ Misses       5062     2199    -2863     
+ Partials     1113      187     -926     

Flag	Coverage Δ
shard-1	42.85% <97.67%> (-9.53%)	⬇️
shard-2	81.70% <23.25%> (+24.59%)	⬆️
shard-3	48.62% <18.60%> (-4.69%)	⬇️

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

Files with missing lines	Coverage Δ
src/review/outcomes-wire.ts	92.51% <100.00%> (+0.92%)	⬆️
src/review/public-stats.ts	97.56% <ø> (ø)
src/review/reversal-superseded.ts	100.00% <100.00%> (ø)
src/services/public-accuracy-trend.ts	100.00% <ø> (ø)

... and 669 files with indirect coverage changes