calibration: per-provider reviewer track records from the consensus corpus

[JSONbored]

Problem

Epic #8211 track F. Dual-reviewer consensus events already exist (reviewer-consensus calibration, packages/loopover-engine/src/reviewer-consensus-calibration.ts), and reversal labels now say which calls were right. Nothing computes what that implies per PROVIDER: measured precision per reviewer identity, per repo, over decided cases.

Requirements

⚠️ Required pattern. Engine-pure extension beside the existing consensus-calibration module — reuse its ingestion shapes; root mirror suite; null-precision-below-sample-floor discipline throughout.

• computeProviderTrackRecords(signals, cases) → per (provider, repo) and per provider overall: {decided, confirmed, reversed, precision|null, split/consensus agreement rates}.
• Deterministic ordering; aggregates only; providers are opaque ids (no config coupling).

Deliverables

• Pure computation, 100% mirror coverage (both-provider/one-provider cases, sample floors, per-repo vs overall rollup invariants)

Links & Resources

#8211 (epic), packages/loopover-engine/src/reviewer-consensus-calibration.ts

Boundaries

Computation only — routing/weighting is the maintainer follow-on. No provider names hardcoded.

[linear-code]

JSO-5712