Normalize and filter dataset

## Summary
Apply normalization rules and filter out low-quality commit messages.

## Success Criteria
- [ ] Apply all normalization rules from foundations
- [ ] Filter criteria defined and applied:
- Remove non-conventional commits
- Remove commits with parsing errors
- Remove duplicates
- Remove auto-generated commits (dependabot, etc.)
- [ ] Quality metrics computed and documented
- [ ] Before/after statistics reported

## Quality Filters
- Minimum subject length
- Valid type (feat, fix, docs, etc.)
- No merge commits
- English language only (v1)

## Output
- Cleaned dataset in canonical schema format
- Quality report with filtering statistics