feat(enrichment): Full-file / near-verbatim duplication scan

[JSONbored]

Context

A REES (review-enrichment service) analyzer. Tier: solid.

Detects: Code added in the PR that is a near-verbatim duplicate of an existing block elsewhere in the repo (copy-paste instead of importing the existing helper), reported as head:line vs source:line.

Data source: GitHub git-tree (recursive) at headSha → fetch same-language files → winnowing/k-gram fingerprint (jscpd/MOSS-style) of added hunks vs the rest of the tree. Pure compute after authed fetches.

This is heavy/external/historical analysis the no-checkout headless claude --print reviewer cannot do; the REES returns it as a brief block the engine splices into the review (additive + fail-safe).

Implementation (established pattern, all inside review-enrichment/)

1. Finding type + BriefFindings key in src/types.ts
2. src/analyzers/<name>.ts — pure, inject fetch for tests
3. Register in src/brief.ts ANALYZERS registry
4. Render a public-safe block in src/render.ts
5. node:test units against dist/ + a live smoke against the real data source

Deliverables

• The analyzer + wiring + tests + a verifiable brief block (file:line or package@version)
• Clean PR off main (zero engine conflict; outside the engine tsc/vitest/codecov scope)

Parent: #1499