Migrate collector semconv codegen to weaver

[jsuereth]

NOTE: This PR is (currently) an example for discussion on the evolution of semconv codegen in the collector.

I regenerated v1.27.0 semconv package so you can get a feel for the changes - I can revert that if we go this route.

Description

Migrate semconv codegen to weaver.

Discovered a few things with codegen in the collector that may be problematic going forward, but did the bare minimum here:

• Migrate from semconvgen docker image (transitive dependency from opentelemetry-go-build-tools). This image is deprecated, will break in the future
• Split out codegen for event, resource + trace. These were not producing attributes today, and going forward I think some time/thought should be made into what this looks like. Semantic conventions has shifted direction overtime, and we have recommendations for codegen that we should co-design a new solution w/ Collector maintainers.
• Update docker calls to run as local user rather than root (meaning generated files are no longer owned by root).
• Remove requirement level from attribute registry (this was never an accurate description and any reliance on it was fundamentally broken).
• Update documentation to leverage automatic git resolution of semconv tags built into weaver.

Link to tracking issue

Tracked at: open-telemetry/weaver#227

Testing

I manually ran the codegen against latest semconv and compared agianst current versions. I can run this codegen in existing directories for clarity.

Documentation

Documentation remains unchanged for codegen. I can add links to weaver documentation if needed.

Open Questions / Concerns

• The Collector, unlike other systems, does NOT fragment stable vs. unstable portions of semantic conventions in codegen, instead having docs outline the current stability of constants. This might be ok given the versioned-packages, want to ensure that's true.
• The existing collector semconvgen template appears to filter on stable attributes, but generated code does not, making me think perhaps the filter wasn't doing what was expected or was broken over time.
• The codegen around events, traces and resource appears to not be producing anything. I think we can improve that, but I'm curious what collector authors/maintainers need here. For now I set up the ability to define a template and preserve existing behavior, but the previous template would not have produced anything useful for these, so I did not duplicate its logic.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 92.20%. Comparing base (631df98) to head (24a46cb).
Report is 108 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #11064      +/-   ##
==========================================
- Coverage   92.23%   92.20%   -0.03%     
==========================================
  Files         405      414       +9     
  Lines       19089    19718     +629     
==========================================
+ Hits        17606    18181     +575     
- Misses       1123     1166      +43     
- Partials      360      371      +11     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[lmolkova]

Weaver/template changes LGTM!

[lmolkova]

requirement level disappeared

[jsuereth]

Yes - There is not such things are requirement level on the attribute registry, that's signal specific.

I'm not even sure where it was pulling requirement level from before, but I call that out in the PR description.

[github-actions]

This PR was marked stale due to lack of activity. It will be closed in 14 days.