fix(server): handle streamed Req telemetry results

What changed

• Normalized the Req/Finch telemetry result values in Tuist.HTTP.PromExPlugin before deriving PromEx request tags.
• Added focused regressions for streamed Req success and streamed timeout/error result shapes.

Why

Sentry was reporting telemetry handler crashes for outbound job-log requests. The PromEx request tag extractor only handled plain results like {:ok, %Req.Response{}} and {:error, exception}. Req telemetry can also report streamed into results as tuple shapes such as {:ok, {%Req.Request{}, %Req.Response{}}} and {:error, exception, {%Req.Request{}, %Req.Response{}}}.

Because those shapes were not matched, the telemetry handler itself crashed with a CaseClauseError, producing issues like https://tuist.sentry.io/issues/127401699/.

Impact

HTTP request metrics continue to be emitted for the streamed Req cases instead of turning telemetry into a production exception source. Successful streamed responses keep their status tag, and streamed failures keep a sanitized error tag.

Validation

• Added the regression first and reproduced the CaseClauseError locally.
• Ran mix format lib/tuist/http/prom_ex_plugin.ex test/tuist/http/prom_ex_plugin_test.exs.
• Ran mix test test/tuist/http/prom_ex_plugin_test.exs.
• Ran git diff --check.