Phase 4: Telemetry, operator overrides, and regression tests

[kai-linux]

Goal

Add observability for dedup behavior and comprehensive regression tests to prevent false-positive/false-negative dedup. Implement operator override mechanics for manual dedup corrections.

Success Criteria

• Daily digest gains "dedup stats" line: N suppressed at groom, M blocked at dispatch, K cross-repo tracked
• Operator override /dedup-reset <issue> re-opens suppressed finding so future signal can file fresh; logged to audit trail
• Regression test: synthetic pair of same-meaning differently-worded issues is deduped by Phase 1
• Regression test: synthetic in-flight task blocks dispatch of same issue in Phase 2
• Regression test: synthetic cross-repo dependency bump produces 1 tracking issue, not N separate issues (Phase 3)
• All regression tests pass in CI

Constraints

• Depends on Phases 1, 2, 3 being complete and integrated
• Prefer minimal diffs — add stats to existing digest template rather than restructuring

Part of #259