fix: accept current vLLM reasoning_effort values

[HuXinjing]

Summary

• accept the reasoning effort values currently supported by vLLM: none, minimal, low, medium, high, xhigh, and max
• preserve the selected value when forwarding requests
• cover both Chat Completions and Responses request schemas with serialization round-trip tests

Motivation

The router currently rejects reasoning_effort: "none" during request deserialization with HTTP 422, before the request can reach a vLLM backend. Current vLLM uses none to disable thinking for supported models and exposes the expanded set of effort values in its OpenAI-compatible protocol.

Validation

• cargo fmt --all -- --check
• cargo test --test test_openai_routing reasoning_effort_accepts_current_vllm_values
• cargo clippy --all-targets --all-features -- -D warnings

Validated separately against a vLLM-backed deployment: a Chat Completions request containing reasoning_effort: "none" returned HTTP 200.

[Copilot]

🟢 Approval recommended

The change is a straightforward schema extension with a focused round-trip test covering all newly supported values.

Pull request overview

This PR updates the router’s OpenAI-compatible request/response schemas to accept the full set of reasoning_effort string values currently used by vLLM (including "none"), and adds a serialization round-trip test to ensure the selected value is preserved when forwarding.

Changes:

• Expand ReasoningEffort to include none, minimal, xhigh, and max in addition to the existing values.
• Update the ChatCompletionRequest doc comment for reasoning_effort to reflect broader vLLM support.
• Add an integration test that validates deserialization + re-serialization preserves each supported value for both Chat Completions and Responses reasoning params.

File summaries

File	Description
src/protocols/spec.rs	Extends ReasoningEffort enum and updates the reasoning_effort field documentation to align with vLLM’s current accepted values.
tests/test_openai_routing.rs	Adds a round-trip serialization test to ensure all supported reasoning_effort values are accepted and preserved.

Review details

• Files reviewed: 2/2 changed files
• Comments generated: 0
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.