feat(commits): semantic commit analysis with risk scoring

Implements issue #32. Provides heuristic analysis of git commit history
without requiring an LLM or external tools:

- Classify: parses Conventional Commits (type, scope, breaking flag)
- ScoreRisk: assigns RiskLow/Medium/High based on diff size, file count,
  breaking changes, reverts, and risk keywords in the message
- FindSimilar: cosine similarity search over pre-computed embeddings
- DetectPatterns: identifies repeated commit types and high-churn files
- Summarize: aggregates risk signals across a set of similar commits

Risk thresholds: score>=3 → High, score>=1 → Medium, else Low.
Breaking changes and reverts always score High.

Co-authored-by: Ona <no-reply@ona.com>

Author: Siddhant-K-code