feat(grafana): add inference and traffic grafana templates for OGL

[tibo-pdn]

Overview

Resolves #766 — Monitoring: add Grafana dashboard templates into the OGL repository.

This PR adds two ready-to-import Grafana dashboard templates under a new grafana/ directory, providing out-of-the-box observability for an OpenGateLLM deployment based on the metrics exposed through Prometheus:

• grafana/inference.json — Inference dashboard (tags: llm, inference, opengatellm). Organized into rows covering Traffic (total requests, request rate, success rate, requests by model, error rate by model & status code), Latency (request duration by model), Time To First Token (TTFT), Tokens (prompt/completion/total tokens and token rates by model) and Output Generation Speed (tokens/s, overall and by model). Includes template variables for datasource, model, endpoint and percentile so panels can be filtered interactively.
• grafana/traffic.json — Traffic dashboard with HTTP-level panels: total requests, request count, average duration, percentage of 2xx / 5xx responses, P99 request duration, request rate, plus Python process resident memory and CPU usage. Includes app_name and DS_PROMETHEUS template variables.

Both dashboards rely on a Prometheus datasource and are importable as-is into any Grafana instance.

DoD criteria for this PR:

• Inference dashboard template added and importable into Grafana
• Traffic dashboard template added and importable into Grafana
• Dashboards use parameterized datasource / template variables (no hardcoded environment-specific values)

Breaking changes:

• No breaking changes
• This PR contains breaking changes (explain below)

Check lists

Review checklist

• Updated or added documentation — N/A: no code documentation changes; the dashboards are self-describing JSON templates.
• Updated or added unit tests — N/A: this PR only adds Grafana dashboard JSON, no application code to test.
• Updated or added integration tests — N/A: no application code changes.
• No debug logs or commented-out code left
• No secrets or environment variables committed in clear text
• Code is linted and formatted using the project pre-commit hooks

api/sql/models.py has not been modified — no Alembic migration required.

Deployment checklist

• Alembic migration has been generated — N/A
• Configuration file has been modified — N/A
• Environment variables have been modified — N/A

No new environment variables and no special deployment steps are required. To use the dashboards, import grafana/inference.json and grafana/traffic.json into Grafana and select the Prometheus datasource that scrapes OpenGateLLM metrics.

Additional Notes

This change is documentation/observability-only: it adds JSON dashboard templates and does not touch any application code, database schema, configuration or runtime behavior. Reviewers may want to confirm that the panel queries match the metric names currently exposed by the OGL Prometheus exporter (request, TTFT, token and output-speed metrics) so the dashboards render correctly against a live instance.