feat: implement execution gap reporting, redundant test detection, CI

[trancee]

Implements three Scott-CC→OMP feature gaps:

T6 — Confidence Intervals

• Wilson score 95% CI formula in wilsonInterval(), clamped to [0,1]
• Added confidenceIntervalLow/High to MutationStats, serialized via encodeDefaults=true
• Score is nullable; CI null when no mutations are evaluable

D3/T7 — Execution Gap Reporting

• ExecutionGap data class (5 gap types: NO_OUTPUT, PARTIAL_RUN, COMPILATION_FAILURE, IR_TRANSFORMATION_ERROR, BACKSTOP_TIMEOUT)
• detectGaps(): parser-level detection for empty stdout (NO_OUTPUT) and footer count mismatch (PARTIAL_RUN)
• Gradle task checks JUnit XML presence → COMPILATION_FAILURE gap when missing
• Score = killed/(total - gaps), null when denominator == 0
• TimedOut is NOT a gap — it's a valid mutation result
• 5 parser tests + 3 serializer tests

D1 — Redundant Test Detection

• RedundantGroup data class (tests, count, failureSignature)
• detectRedundantTestGroups(): groups tests by failure signature, threshold >5, sorted by count desc, excludes zombies
• 5 parser tests

D2 — Modes, Focus, Auto-Approve (agent-level docs)

• --mode quick|standard|deep → mutflow maxRuns (10/30/all)
• --focus → Gradle includeTargets/excludeTargets bridge
• --auto-approve gate: zombie/redundant deletion requires explicit approval
• Updated SKILL.md, test-quality-reviewer.md, test-refactor-specialist.md, test-auditor.md, test-executor.md

Tests

35 tests (18 parser + 7 serializer + 10 stats), all passing.