Question about minimal API codebase change: Making openAIAdapterStream protected (now private)

[dixoxib]

File in question: core/llm/index.ts
Change: private openAIAdapterStream → protected openAIAdapterStream
Reason: Need access to raw API finish_reason for full model feature integration

Currently there's no way to access finish_reason from OpenAI-compatible adapters at the LLM implementation level. The cleanest solution for provider-specific streaming logic seems to be overriding openAIAdapterStream.

Any objections to this visibility change? If yes, what's the preferred approach for implementing provider-specific streaming logic that needs raw API response data?