Log responses mode and coerce developer role

Ignore penalty params in requests

Align responses fallback with OpenAI output

feat: model-based adapters for chat/responses

feat: add responses adapter for OpenAI types

refactor: passthrough openai new-style streaming

Fix streaming multi-choice and legacy defaults

fix: broaden OpenAI compatibility for inputs and outputs

fix: align legacy chat.completions response

fix: disable streaming when client doesn\'t accept SSE

fix: stabilize responses fallback streaming

fix: add response output item event for chat fallback

fix: normalize tools fallback and log params

Remove tiktoken and simplify Docker build

Fix Docker build deps file generation