Emit output_item.done, reasoning_summary_part.done, function_call_arguments.done,
and related finalize events for text, reasoning, and tool calls before the
response.completed event to match the Responses API stream contract.
* feat(aiproxy): support streaming responses for visual messages
* refactor(aiproxy): share API log helpers and support input/output token usage
Extract common chatlog record lifecycle across proxy handlers, and fall back
to Responses/Anthropic usage field aliases when filling token counts.
Promote visual_provider_id/visual_model_key to dedicated ai_model columns,
force non-stream upstream orchestration for visual paths, and synthesize SSE
chunks for streaming Responses requests that include images.
* refactor(aiproxy): rename chat log config to API log with S3 fields
* feat(aiproxy): add visual extension for Responses image tools
Wire visual model catalog, orchestration, and OpenAI Responses compat so chat/Codex can invoke image generation and editing tools.
Add /v1/messages handler with Anthropic-to-OpenAI translation, upstream
failover, and probe endpoints. Replace shell-based functional test scripts
with pkg/aiproxy/ft and climc test commands; consolidate documentation.
Add auto_register_aiproxy on llm_deployment creation, background sync task,
and register/unregister CLI actions. Link aiproxy providers, routings, and
models to llm replicas via llm_id and llm_deployment_id fields.
Zexi Li