[grpc] Refactor openai module

[CatherineSue]

Motivation

openai router module has the following problems:

1. route_responses() is 270 lines of inline logic (830-1097)

route_responses() in router.rs:
├── Metrics recording
├── Worker selection
├── Previous response chain loading (~30 lines)
├── Conversation history loading (~90 lines)
├── Payload transformation
├── Context building
└── Finally: dispatch to streaming/non-streaming

Compare to gRPC's route_responses_impl() (~50 lines):

// gRPC just validates and delegates
validate_worker_availability(&self.worker_registry, model);
if is_harmony {
    serve_harmony_responses(&harmony_ctx, body.clone()).await
} else {
    responses::route_responses(&self.responses_context, ...).await
}

2. handle_non_streaming_response() is inline in router.rs (371-491)

This 120-line function should be in a separate file like Harmony's non_streaming.rs.

3. Circular dependency between files:

accumulator.rs ──imports──> streaming.rs (extract_output_index, get_event_type)

These helpers should be in a common utilities file, not streaming.rs.

4. Mixed responsibilities:

File	What it has	What it SHOULD have
router.rs	Route + non-streaming handler + conversation loading	Just routing (like gRPC)
responses.rs	Tiny utils only (225 lines)	Could be the entry point
streaming.rs	Handlers + SSE helpers	Just handlers
mcp.rs	Non-streaming loop + streaming execution	Tool execution only

Modifications

  1. Moved files from openai/ into openai/responses/ subdirectory
  2. Extracted handle_non_streaming_response() from router.rs to responses/non_streaming.rs
  3. Created common.rs with shared SSE helpers used by both streaming and accumulator
  4. Removed unused re-exports of Provider, ProviderError, ProviderRegistry

Accuracy Tests

Benchmarking and Profiling

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
• Follow the SGLang code style guidance.

Review Process

1. Ping Merge Oncalls to start the PR flow. See the PR Merge Process.
2. Get approvals from CODEOWNERS and other reviewers.
3. Trigger CI tests with comments (/tag-run-ci-label, /rerun-failed-ci, /tag-and-rerun-ci) or contact authorized users to do so.
4. After green CI and required approvals, ask Merge Oncalls to merge.

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!