Improve OpenAI evaluator telemetry coverage with gpt-5.5 xhigh (plan …

[mltseva]

artifact_path: experiments/tracy-openai/gpt-5.5-xhigh-medium/
Eval id: 294156bb-e4d9-4f1e-9c9f-f211e9b9ae9a
Status: succeeded
Score: 98
Scenarios: 67/86 passed, 19 partially failed.

Main remaining gaps:

• openai/chat/lifecycle: largest loss, score 31. create passes, but retrieve returns 404, so later update/messages/list/delete spans are missing.
• openai/chat/vision: API returns 400 invalid_image_url; also missing request-side tracy.request.input_image.count.
• openai/chat/streaming: missing gen_ai.request.stream=true, response id/model, and finish reasons.
• openai/images/streaming and openai/images/edit_streaming: gen_ai.request.stream is emitted as string "true" instead of Boolean true; missing tracy.response.created_at.
• openai/images/variation: response is 404, operation is generate_content, and image request/response attrs are missing.
• Smaller gaps: embeddings dimension count, conversations item retrieve operation name, Responses file/image input attrs, tracy.request.include, invalid input error type/code, and videos delete response attrs.

Highest-value fixes: boolean serialization for gen_ai.request.stream, embeddings dimension count, correct conversations retrieve operation, request-side multimodal/file attrs, and image variation route handling.