Phase 3: Cross-repo dedup for shared concerns

[kai-linux]

Goal

Extend dedup to be cross-repo for inherently shared concerns like "bump X dependency" or "propagate new audit-trail usage". Instead of filing N identical issues (one per affected repo), file a single tracking issue and attach per-repo concerns as comments.

Success Criteria

• Groomer checks cross-repo when finding carries cross_repo_scope: true hint (set by detector)
• Match by stable concern_id (e.g., dep-bump:sqlalchemy:2.0.x) using semantic dedup infrastructure from Phase 1
• If same logical concern already has open issue in another repo, new finding is attached as comment to tracking issue rather than filed separately
• Operator override available via /dedup-split <issue> if per-repo work is actually needed
• Integration tested: verify that dependency bump across 3 repos produces 1 tracking issue with 3 comments, not 3 separate issues

Constraints

• Depends on Phase 1 semantic dedup infrastructure (concern_id matching, similarity scoring)
• Prefer minimal diffs — extend Phase 1 logic rather than forking dedup code

Part of #259