[Http Metrics] Replace http.error.reason with OTel standard error.type

[antonfirsov]

Align attribute name and semantics with the OpenTelemetry spec (which was finalized yesterday 2023/9/11 - see open-telemetry/semantic-conventions#205) -- http.error.reason attribute has been standardized under name error.type. We need to:

• Rename http.error.reason to error.type.
• Report error.type also for valid responses where the HTTP status code indicates an error (4xx and 5xx) -- in such case, the value should be the string representation of the HTTP status code.
  • Note: Currently we report http.error.reason only when the underlying handler fails to fetch a response.

See error.type in the spec for more details.

For context: We introduced the attribute http.error.reason in PR #89809 (on 2023/8/3) based on the in-progress draft of the OTel spec (open-telemetry/semantic-conventions#205).

Given that this is new 8.0 feature, we should adapt to the spec in 8.0 to avoid breaking changes with 9.0.

[ghost]

Tagging subscribers to this area: @dotnet/ncl
See info in area-owners.md if you want to be subscribed.

Issue Details

---

The http.error.reason attribute implemented in #89809 has been standardized as error.type with minor changes to its' semantics in
open-telemetry/semantic-conventions#205. To avoid a breaking change or the introduction of a noisy secondary attribute in future releases, we should consider adapting the standard attribute naming and semantics in 8.0.

This means that:

• We should rename http.error.reason to error.type
• Currently error.type is reported only in cases when the underlying handler fails to fetch a response. With the new spec we should also report error.type when there is a valid response and the response status code indicates an error. In this case error.type should be the string representation of the status code.

See error.type in the spec for more details.

Author:	antonfirsov
Assignees:	-
Labels:	area-System.Net.Http
Milestone:	8.0.0

[karelz]

Reopening to track backport to 8.0

[karelz]

Fixed in main (9.0) in PR #91910 and in 8.0 (RC2) in PR #91996