fix(tracing): name streamed task spans after the run's own workflow

[LeSingh1]

Summary

A task span represents one top-level Runner invocation. The non-streaming path names it from the workflow name resolved for that run; the streaming path named it from streamed_result.trace or get_current_trace().

Those two agree only when the run creates its own trace. Inside an existing trace create_trace_for_run returns None, so streamed_result.trace is None and the streamed run fell back to the enclosing trace's name. An orchestration trace that fans out several streamed sub-runs therefore gets task spans all labelled with the orchestrator's workflow name — indistinguishable from each other and from the outer trace — while the same code run non-streamed labels them correctly. RunConfig(workflow_name=...) is effectively ignored for streamed runs nested in a trace.

This passes the already-resolved workflow name into start_streaming and uses it, which also drops a current_trace guard the non-streaming path never had (that variable existed only to supply the name, and its None branch is unreachable — create_trace_for_run returns None only when get_current_trace() is truthy).

start_streaming lives in agents.run_internal and is not a public export, so the new keyword-only argument is not a public API change. No persisted schema or wire format is affected; only exported telemetry changes, and it changes toward the documented semantics and toward the non-streamed path.

Present in released v0.19.2.

Test plan

• New tests/test_agent_tracing.py::test_wrapped_run_task_span_uses_run_workflow_name runs the same agent non-streamed and streamed inside one outer trace and asserts both task spans use the run's own workflow name. It fails on main with At index 1 diff: 'outer_workflow' != 'inner_workflow' and passes with this change.
• uv run pytest tests/test_agent_tracing.py tests/tracing → 72 passed.
• bash .agents/skills/code-change-verification/scripts/run.sh → all commands passed.

Issue number

N/A — found while auditing tracing parity between the streamed and non-streamed runners.

Checks

• I've added new tests, if relevant
• I've run .agents/skills/code-change-verification/scripts/run.sh
• I've confirmed all verification steps pass
• If using Codex, I've run /review before submitting this PR