Auto-instrumentation and wrapper traces differ for Google GenAI

[lforst]

Summary

Google GenAI auto-instrumentation and wrapper instrumentation both capture the same operations, but they do not emit the same trace shape.

What differs

The auto-instrumented path emits different span names / metadata shape than the wrapper path for the same Google GenAI calls.

Why this matters

This makes it harder to share one strict trace contract across wrapper and auto-instrumentation tests, and it means users may see different trace structure depending on how instrumentation is enabled.

Repro

The aligned e2e coverage in:

• e2e/scenarios/wrap-google-genai-content-traces
• e2e/scenarios/google-genai-auto-instrumentation-node-hook

shows the same operations being traced with different emitted span shape.

Expected

Wrapper and auto-instrumentation should converge on the same trace contract for Google GenAI where possible, especially for span naming and logged metadata.