feat: implement M2 — CI failure parsing and AI review generation

[shenxianpeng]

Summary

Implement M2 (Q4 2026): CI Failure & AI Review — demonstrating the AI-native value proposition.

Changes

Module 04 — AI Change Review (Draft → Candidate)

• ods review generate: Full L1/L2/L3 record generation
  • --level L1|L2|L3 flag with auto-detection from AI percentage
  • Correct checklist per level (L1: correctness/quality, L2: +security/ai_specific, L3: +architecture/documentation/compliance)
  • L3 includes second_reviewer requirement
• ods review validate: Validates against ODS AI Change Review schema
• ods review ai-percentage: Estimates AI contribution from commit log footers via stdin

Module 05 — CI Failure (Draft → Candidate)

• ods ci parse: Full CI log parser with AI hallucination detection
  • Detects 6 hallucination categories: non-existent symbols, wrong imports, deprecated APIs, fake URLs, incorrect defaults, hallucinated function calls
  • Produces structured JSON reports with confidence levels
  • Suggests closest valid symbols via Levenshtein distance
• ods ci explain: Human-readable failure analysis highlighting AI-related issues
• ods ci fix-suggestions: Prioritized fix list with auto-fix availability

New: internal/ciparser package

• Complete CI log parsing with stage detection, failure classification, duration tracking
• AI hallucination detection engine with regex patterns for each category
• Fix suggestion generation with deduplication

Testing

# Review generation
go run ./cmd/ods review generate --pr 42 --level L2 --ai-pct 45
go run ./cmd/ods review generate --pr 99 --level L3 --ai-pct 92

# CI parsing with hallucination detection
echo 'FAIL TestBuildQuery — hallucinated function call: buildQuery() does not exist
FAIL compile error: undefined symbol refreshTokenV2 in auth/token.go' |   go run ./cmd/ods ci parse --stdin --pipeline build-42 --repo org/svc

Type

• Feature