Add request metrics hooks #121

Description

Expose request-count, latency and error observations so applications can track volume, latency percentiles and error rate.

Current middleware records formatted request logs; no public metrics hook/observer abstraction is exported. KV timing logs measure store operations, not request metric aggregation.

Steps to reproduce

1. Inspect core middleware and module exports for a metrics observer.
2. Compare request log output with count, duration-distribution and error observations needed by a metrics backend.

Expected behavior

Applications can attach a metrics backend to reusable request observations without replacing request handling.

Impact

Applications must implement timing/count/error hooks independently before exporting request metrics.

Proposed fix

Define a portable request-observation hook with bounded labels and explicit completion/error boundaries; leave backend aggregation/export and percentiles to the selected metrics integration.

Done when

• Hooks expose request count, duration and outcome/error facts.
• Route/method/status labels and unmatched/error behavior are defined.
• A backend example derives request rate, error rate and latency percentiles.
• Hooks respect the portable runtime boundary and avoid unbounded per-request label state.

Related

• Parent: Complete portable request observability #118.
• Timing collection: Expose request latency automatically #120 and Add a shared request-timing collector and opt-in attachment middleware #446.
• Native diagnostics: Axum: add safe production diagnostics and trusted-proxy handling #397.
• Built-in request logger
• Core public module surface
• Related PR feat(core): add shared request timing collector and middleware #389; this PR alone does not establish completion of this issue.
• Related PR feat(axum): add safe diagnostics and trusted proxy metadata #409; this PR alone does not establish completion of this issue.