fix(python): update profile expectations

[taegyunkim]

Summary

• update Python stack expectations for module-qualified time.sleep and gevent frames
• disable tracing-only background services in the Python profiler test base image

Fixes the failures from https://github.com/DataDog/prof-correctness/actions/runs/31654920913.

Why

These scenarios install unpinned dependencies. Run 31496036887 passed on the same prof-correctness commit shortly before dd-trace-py v4.13.0 was released, while the next scheduled runs pulled the new release and exposed two issues.

• Native frame names: dd-trace-py #19069, shipped in v4.13.0, changed native-frame display names from the bare function name to <module>.<function>. The observed stacks therefore changed from sleep to time.sleep and from loop.run to gevent.libev.corecext.loop.run; the workload and expected durations did not change.
• Unrelated background threads: wall-time percentages use every profiled thread as the denominator. Code Origin's SignalUploader and the remote-config poller can appear as long-lived Dummy-1 threads. Greenlet's two branches consequently varied from 29-33% when a background thread was present to 48-49% when it was not. There is no single correct percentage for that unstable denominator.

The Python base image already disables tracing because these scenarios test the profiler. This change also disables Code Origin and remote configuration, removing their unrelated background threads. The original duration-independent percentage targets and error margins remain unchanged.

With dd-trace-py v4.13.0 and both background services disabled locally, greenlet's branches measured 47% and 49% against the original 50% ± 15% expectations. uWSGI measured 47% for the requester and 45% for the worker against the original 50% ± 20% and 42% ± 10% expectations.

The other scenarios named in the Slack messages passed. They were only grouped into the same failed matrix shards.

Testing

• ran the greenlet and uWSGI containers locally with dd-trace-py v4.13.0 and the updated base environment, then analyzed their generated profiles against the original relative expectations
• replayed the stack-name expectations against artifacts from run 31654920913
• go test ./analysis/... ./cmd/list-scenarios/... ./cmd/prof-analyze/...
• go test -run '^TestSchemaValidation$' -count=1

[taegyunkim]

@KowalskiThomas Just turned off code origins and remote config which simplified things a bit

[KowalskiThomas]

Nice find with the other threads -- note sure why we had to disable CO for Spans for that to work (RC makes sense, CO for Spans less so), but in any case LGTM.

[taegyunkim]

/code blockers

[gh-worker-devflow-routing-ef8351]

View all feedbacks in Devflow UI.

2026-08-14 14:13:52 UTC ℹ️ Start processing command /code blockers

---

2026-08-14 14:13:53 UTC ℹ️ Devflow:

Checking merge blockers for #182...

---

2026-08-14 14:13:59 UTC ℹ️ Devflow: /code blockers

Detected 1 merge blocker(s) to address:

🔴 Blocking

• This pull request contains unsigned commits.
  Commit 1f8f9e1 is not signed.
  Hint: Sign your commits, then force-push. See our FAQ.