feat: log backend response telemetry

[phaedrus-on-git]

Summary

• log completed backend responses with alias, target model, upstream model, status, duration, bytes, and OpenAI usage token counts when present
• log backend proxy errors with alias/model and elapsed duration
• add token usage to the mock backend and a unit test for telemetry logging

Validation

• go test ./...
• go run ./cmd/devrail-router check -config configs/router.example.yaml
• make package-smoke GOOS=linux GOARCH=amd64
• make docker-smoke
• make check