Datadog: define a metrics path, then ship a dashboard template

Description

Roughly 50 aisix_* Prometheus series are exposed on a dedicated metrics listener, and there is a native Datadog exporter — but it targets the Logs HTTP intake only (crates/aisix-obs/src/sink/datadog.rs), so there is no Datadog metrics/series path at all. observability.metrics.otlp parses into OtlpConfig but has no consumers; OTLP metrics export is still on the roadmap.

The repo also ships no importable dashboard template for any backend, and the docs contain no mapping from the aisix_* series to Datadog.

Requested, in order:

1. Decide and implement how aisix_* metrics reach Datadog — Datadog metrics intake, Agent/OpenMetrics scrape of the metrics listener, log-based metrics off the existing intake, or finishing the OTLP metrics stub.
2. Author a Datadog dashboard template, importable from the repo, covering requests and errors by model, latency percentiles, token throughput, cost, cache hit rate, guardrail blocks and rate-limit rejects.

Why

Teams standardised on Datadog judge an integration by whether metrics arrive and a dashboard exists to import; a logs-only exporter (#57, api7/AISIX-Cloud#688) does not give them a gateway view.

Prior art exists in the portfolio — the API7 EE datadog plugin ships metrics via DogStatsD.

Priority

Low.

Prior art

LiteLLM is the only peer covering both halves, and has an official Datadog integration tile.

_{Surveyed 2026-07-15; every claim rests on a fetched docs/source page.}