### Check for existing issues

- [x] I have searched the existing issues and checked that my issue is not a duplicate.

### What happened?

Non-streaming completions via `github_copilot/*` correctly surface real extended-thinking output in `message.provider_specific_fields.reasoning_text` (+ `reasoning_opaque` signature) when using `thinking: {"type": "adaptive"}` for adaptive-thinking-capable models (`claude-sonnet-5`, `claude-opus-4.6/4.7/4.8`, `claude-sonnet-4.6`). The identical request with `stream=True` returns **zero** reasoning data — only regular content deltas — even though GitHub Copilot's raw API does stream reasoning for this exact request (verified via raw HTTP, bypassing litellm entirely: the SSE contains `delta.reasoning_text` per chunk and a final `delta.reasoning_opaque` chunk).

**Root cause:** `GithubCopilotConfig` (`litellm/llms/github_copilot/chat/transformation.py`) extends `OpenAIConfig` and doesn't override streaming, so it falls back to the generic `OpenAIChatCompletionStreamingHandler.chunk_parser()`, which just does `ModelResponseStream(**chunk)`. That handler has no special-case mapping for Copilot's `reasoning_text`/`reasoning_opaque` keys, so Pydantic …