Add cache-hit-specific duration metric (contrast with cache_resolution)

The existing `duration` metric can't easily be filtered to show only cache hits, making it hard to contrast hit vs. miss performance against `cache_resolution`.

## Ask

Either:
1. Add a `cache_hit_duration` metric that captures latency for requests served from cache, OR
2. Add a `cache_hit` label/dimension to the existing `duration` metric so it can be filtered in Grafana.

The goal is to be able to show: "cache hit latency = X ms vs. cache miss latency = Y ms" in a dashboard.

If there's already a way to compute this with existing metrics, documentation would be equally valuable.

---

🤖 Filed by Claude on behalf of Kris.