Decide the chat-to-Responses bridge from the provider's Responses config, so a
responses-mode model on a provider without one stays on chat instead of bouncing
through responses() and back. That bounce was the only loop
_skip_responses_api_bridge guarded, so the flag goes away.
The /v1/messages bridge for chat-only models now plans the chat call once
and runs execute_chat directly instead of re-entering litellm.acompletion,
so deployment hooks run once per request. Requests that need the chat
surface (fallbacks, model_list, MCP gateway tools, prompt management,
litellm_system_prompt) still go through litellm.acompletion
Messages, chat and Responses each decided inline whether to bridge to
another API format. The decisions now live in litellm/native/route_plan.py
as a tagged union so the upcoming executor split has one place to read