fix(server): Measure request duration end-to-end

[lcian]

server.requests.duration now measures the request through response-body completion rather than stopping when a handler produces headers. This includes time spent streaming GET payloads, batch responses, and multipart-complete responses.

The middleware moves its timing guard into MetricsBody, which emits the metric only after Hyper has completed or dropped the response body. The wrapper preserves response metadata and existing Content-Length behavior.

Status tagging reflects how streaming ended:

• successful completion: the response status from the headers;
• client disconnect before completion: 499;
• server-side stream error: 500.

Completion handling is verified against Hyper 1.10.1 for empty/no-body responses, buffered bodies, ordinary streamed bodies, and terminal trailers. Regression tests cover each completion path, plus stream errors and interrupted streams.

Close FS-555

[jan-auer]

The server.requests.duration metric now records the full request lifetime

Changing .duration to cover the full time span makes sense. It can be useful to retain the old TTFB metric in addition to the full duration. We indirectly have those measurements through the server calls, but having a metric that covers all requests is more convenient.

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit c1d2a36. Configure here.}