Recommend model routing from evaluation results

## Summary

Add explainable model-routing recommendations based on Flowly benchmark reports. This is an optional follow-up to #38.

## Scope

- Recommend models by workload type and configured quality, cost, and latency constraints
- Use compatible persisted reports and versioned suite lineage
- Explain the metrics and trade-offs behind each recommendation
- Fail closed when evidence is missing, stale, or not comparable

## Acceptance criteria

- [ ] Recommendations can target repository questions, issue work, PR reviews, and coding tasks
- [ ] Operators can set explicit quality, cost, and latency constraints
- [ ] Every recommendation cites the reports, suite lineage, and metrics used
- [ ] Incompatible or insufficient runs do not produce a confident recommendation
- [ ] Deterministic tests and documentation cover ranking and no-recommendation cases

## Non-goals

- Automatic provider credential changes
- Silent production routing changes
- Dashboard UI or hosted report sharing

Start by reviewing follow-up issue #38 and the existing persisted Flowly benchmark reports and versioned suite lineage. Define how workload, quality, cost, and latency constraints feed ranking, then verify deterministic tests and documentation cover explainable recommendations and no-recommendation cases.

Written by the indexing model from the issue text.