OpenAI Chat Completions streaming drops reasoning (`delta.reasoning_content` is never emitted as a StreamFrame)

[adamglin0]

Summary

When streaming via the OpenAI Chat Completions endpoint, reasoning content is
silently dropped. Providers that return chain-of-thought in the streaming delta as
choices[].delta.reasoning_content (e.g. DeepSeek and OpenAI-compatible gateways)
produce no StreamFrame.ReasoningDelta / StreamFrame.ReasoningComplete, so the
reasoning is lost both during streaming and in the final toMessageResponse().

The non-streaming path works correctly — AbstractOpenAILLMClient reads
reasoning_content and produces a MessagePart.Reasoning — so switching a model
from execute() to executeStreaming() causes reasoning to disappear.