fix: Gemini tool_call_id 跨 content 配对 + Ollama 请求侧 assistant 推理回放

[HandSonic]

📝 变更描述 / Description

修复协议适配层两个独立缺陷（均可在单元测试级别 100% 复现，详见 #6602 与 #6603）：

1. Gemini→Chat 转换 tool_call_id 配对失败（relaykit/relayconvert/internal/gemini_chat/to_oai_chat_req.go）

原实现中 functionCall 的 id 从当前 content 的 toolCalls 切片长度推导（call_1 起），而 functionResponse 属于下一个 content 迭代、切片恒为空，导致 tool 消息的 tool_call_id 恒为 call_0 —— tool_calls 与 tool 消息任何情况下都配对失败，严格校验配对的 OpenAI 兼容上游会直接报错；该错误值已被 golden 快照固化。

修复方式：id 改为请求级递增序号分配；functionResponse 按 Gemini 协议的配对键（函数名）与先前的 functionCall 做 FIFO 配对；无法匹配的 response 分配不冲突的新 id。新增 4 组单测（单调用跨 content 配对、并行调用按名配对、同名重复调用 FIFO、未匹配回退），并重生 3 个 gemini_to_* golden 快照——diff 仅为响应位置 call_0→call_1，证明无其他行为漂移。

2. Ollama 渠道请求侧不回放 assistant 推理（relay/channel/ollama/relay-ollama.go）

openAIChatToOllamaChat 构建 OllamaChatMessage 时映射了 role/content/images/tool_calls，但从不读取 GetReasoningContent()、从不设置 Thinking 字段（响应侧已有上游 thinking → reasoning_content 的反向映射，即"拿得到、传不回"）。

修复方式：assistant 消息的 GetReasoningContent() 映射到 OllamaChatMessage.Thinking（仅 assistant 角色），与响应侧形成对称往返。新增单测覆盖 assistant 映射与非 assistant 角色不映射。

🚀 变更类型 / Type of change

• 🐛 Bug 修复 (Bug fix)

🔗 关联任务 / Related Issue

• Closes fix: Gemini→Chat 转换 tool_call_id 配对失败 #6602
• Closes fix: Ollama 渠道请求侧不回放 assistant 推理（reasoning_content → Thinking 未映射） #6603

✅ 提交前检查项 / Checklist

• 人工确认: 我已亲自整理并撰写此描述，没有直接粘贴未经处理的 AI 输出。
• 非重复提交: 我已搜索现有的 Issues 与 PRs，确认不是重复提交。
• Bug fix 说明: 已关联 Issue fix: Gemini→Chat 转换 tool_call_id 配对失败 #6602 与 fix: Ollama 渠道请求侧不回放 assistant 推理（reasoning_content → Thinking 未映射） #6603，问题可在转换层/渠道适配层单元测试稳定复现。
• 变更理解: 我已理解这些更改的工作原理及可能影响。
• 范围聚焦: 本 PR 未包含任何与当前任务无关的代码改动。
• 本地验证: 已在本地运行并通过测试（见下方运行证明）。
• 安全合规: 代码中无敏感凭据，且符合项目代码规范。

声明：实现、测试与本描述由 AI 辅助完成，已经人工核对。

📸 运行证明 / Proof of Work

新增测试

relaykit/relayconvert/internal/gemini_chat: TestGeminiGenerateContentRequestToOpenAIChatPairsToolCallIDs
  （单调用跨 content 配对 / 并行调用按名配对（乱序响应）/ 同名重复调用 FIFO / 未匹配回退不冲突）
relay/channel/ollama: TestOpenAIChatToOllamaChatPreservesAssistantReasoning
  （assistant 映射 / user 角色不映射）

构建与测试

cd relaykit && GOWORK=off go build ./...          # 通过（模块独立性验证）
cd relaykit && GOWORK=off go test -count=1 ./...  # 全部通过
go build ./relay/... ./service/... ./controller/... # 根模块通过
go test ./relay/channel/... ./service/...          # 全部通过（service 包
  # TestObserveChannelAffinityUsageCacheByRelayFormat 在 origin/main 基线上同样失败，
  # 为上游既有问题，与本 PR 无关）

Golden 快照：3 个 gemini_to_* 请求快照更新，diff 仅为 tool 结果/响应位置的 call_0→call_1（与 assistant 侧既有的 call_1 配对），证明行为变更严格收敛在 id 配对上。

对抗审查：3 视角独立审查（配对算法逻辑攻击、无工具/无推理流量零回归验证、协议形状核对），全部通过。

Summary by CodeRabbit

• Bug Fixes

  • Preserved assistant reasoning across multi-turn Ollama conversations involving tool calls.
  • Improved Gemini-to-OpenAI conversion by reliably matching tool responses, including repeated or parallel calls.
  • Preserved the correct ordering of assistant content and tool responses.
  • Assigned unique identifiers to unmatched tool responses to prevent conflicts.

• Tests

  • Added coverage for reasoning preservation, tool-call matching, fallback identifiers, and message ordering.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: dd55ec8f-9816-4c2d-b4fe-ccbbf84bec9e

📥 Commits

Reviewing files that changed from the base of the PR and between 09ff415 and 7abe706.

📒 Files selected for processing (3)

• relay/channel/ollama/relay-ollama.go
• relay/channel/ollama/relay_ollama_req_test.go
• relaykit/relayconvert/internal/gemini_chat/to_oai_chat_req_test.go

🚧 Files skipped from review as they are similar to previous changes (3)

• relay/channel/ollama/relay_ollama_req_test.go
• relay/channel/ollama/relay-ollama.go
• relaykit/relayconvert/internal/gemini_chat/to_oai_chat_req_test.go

---

Walkthrough

The request conversion changes preserve assistant reasoning for Ollama and coordinate Gemini tool-call identifiers across calls and responses. Tests and golden fixtures validate pairing, ordering, fallback identifiers, and reasoning-field mapping.

Changes

Ollama reasoning replay

Layer / File(s)	Summary
Assistant reasoning mapping relay/channel/ollama/relay-ollama.go, relay/channel/ollama/relay_ollama_req_test.go	Assistant reasoning is serialized into Ollama Thinking. User reasoning is ignored. Serialization failures leave Thinking unset. Tests verify the mapping.

Gemini tool-call pairing

Layer / File(s)	Summary
Tool-call sequencing and ordering relaykit/relayconvert/internal/gemini_chat/to_oai_chat_req.go, relaykit/relayconvert/internal/gemini_chat/to_oai_chat_req_test.go, relaykit/relayconvert/testdata/golden/request/*	Tool calls receive request-wide sequential IDs. Responses consume matching IDs by function name in FIFO order. Unmatched responses receive unique fallback IDs. Buffered tool messages follow their assistant content. Tests and golden fixtures validate pairing and ordering.

Estimated code review effort: 3 (Moderate) | ~25 minutes

Possibly related issues

• fix: Gemini→Chat tool_call_id 配对失败 + Ollama 请求侧不回放 assistant 推理 #6601 — Covers both Gemini tool-call ID pairing and Ollama assistant reasoning replay addressed by this change.
• fix: Gemini→Chat 转换 tool_call_id 配对失败 #6602 — Tracks the Gemini tool_call_id pairing defect fixed by the conversion changes.
• fix: Ollama 渠道请求侧不回放 assistant 推理（reasoning_content → Thinking 未映射） #6603 — Tracks the Ollama assistant reasoning mapping defect fixed by the request conversion.

Possibly related PRs

• QuantumNous/new-api#6395 — Preserves assistant reasoning across related multi-turn tool-call conversions.
• QuantumNous/new-api#6593 — Modifies request conversion paths to preserve reasoning across protocol and tool-call turns.
• QuantumNous/new-api#6605 — Also modifies Ollama request conversion to preserve assistant reasoning in Thinking.

Poem

A rabbit checks each call in line,
And keeps the reasoning aligned.
Ollama stores the thoughts once more,
Gemini pairs each tool-call score.
The tests confirm the IDs agree.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly summarizes both primary fixes: Gemini tool-call pairing and Ollama assistant reasoning replay.
Linked Issues check	✅ Passed	The changes satisfy #6602 and #6603, including FIFO tool-call pairing, fallback IDs, assistant-only reasoning mapping, and focused tests.
Out of Scope Changes check	✅ Passed	All code, tests, and golden fixture updates directly support the two linked issue objectives.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@relaykit/relayconvert/internal/gemini_chat/to_oai_chat_req_test.go`:
- Around line 125-131: Add require.Len(t, toolIDs, 2) immediately after
obtaining toolIDs in the test, before the indexed assertions, so toolIDs[0] and
toolIDs[1] are accessed only after confirming both elements exist.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: fc6d85a7-d705-459e-bd2b-711f6ddd54cc

📥 Commits

Reviewing files that changed from the base of the PR and between 0ab0202 and 09ff415.

📒 Files selected for processing (7)

• relay/channel/ollama/relay-ollama.go
• relay/channel/ollama/relay_ollama_req_test.go
• relaykit/relayconvert/internal/gemini_chat/to_oai_chat_req.go
• relaykit/relayconvert/internal/gemini_chat/to_oai_chat_req_test.go
• relaykit/relayconvert/testdata/golden/request/gemini_to_claude.golden.json
• relaykit/relayconvert/testdata/golden/request/gemini_to_openai.golden.json
• relaykit/relayconvert/testdata/golden/request/gemini_to_openai_responses.golden.json