feat: add runtime metrics reports for LLM, Pilot, and Retrieval

- Add LlmMetricsReport with total calls, token usage, success rates,
  latency, and estimated costs
- Add PilotMetricsReport with decision accuracy, guidance calls,
  and intervention tracking
- Add RetrievalMetricsReport with query performance, cache metrics,
  and path quality statistics
- Add combined MetricsReport to aggregate all subsystem metrics
- Implement metrics hub for centralized metric collection
- Expose metrics_report() method in Python bindings

Author: zTgx