reasoning_content lost during round-trip in tool calling with deepseek-reasoner

Problem

When using deepseek-reasoner model with tool calling, the reasoning_content field is lost during the conversation round-trip, causing the DeepSeek API to return HTTP 400:

Missing `reasoning_content` field in the assistant message

Background

The DeepSeek API requires that every assistant message in the conversation history includes the reasoning_content field when using deepseek-reasoner. This is documented at: https://api-docs.deepseek.com/guides/thinking_with_tools

The typical tool calling flow is:

1. Send user message with tools -> API returns assistant message with reasoning_content + tool_calls
2. Send back the full history (user msg + assistant msg + tool result) -> API expects reasoning_content in the assistant message

Root Cause

The reasoning_content field is correctly parsed from the API response into ContentChoice.ReasoningContent (added in PR #1121), but it cannot be sent back in the next request because:

1. MessageContent struct lacks ReasoningContent field (llms/generatecontent.go) -- the struct only has Role and Parts, so there is no way to attach reasoning_content when constructing assistant messages for conversation history.

2. Conversion MessageContent -> ChatMessage doesn't transfer ReasoningContent (llms/openai/openaillm.go) -- even if the field existed, the conversion code for ChatMessageTypeAI only sets msg.Role = RoleAssistant without copying reasoning content.

Note: The internal ChatMessage struct already has the ReasoningContent field and its MarshalJSON correctly serializes it. The gap is only in the public-facing MessageContent struct and the conversion layer.

Reproduction

llm, _ := openai.New(
    openai.WithModel("deepseek-reasoner"),
    openai.WithToken(apiKey),
    openai.WithBaseURL("https://api.deepseek.com/v1"),
)

tools := []llms.Tool{{
    Type: "function",
    Function: &llms.FunctionDefinition{
        Name:        "calculator",
        Description: "Add two numbers together",
        Parameters: map[string]any{
            "type": "object",
            "properties": map[string]any{
                "a": map[string]any{"type": "number"},
                "b": map[string]any{"type": "number"},
            },
            "required": []string{"a", "b"},
        },
    },
}}

// Step 1: works fine - API returns reasoning_content + tool_call
resp, _ := llm.GenerateContent(ctx, []llms.MessageContent{
    {Role: llms.ChatMessageTypeHuman, Parts: []llms.ContentPart{
        llms.TextContent{Text: "What is 15 + 28? Use the calculator tool."},
    }},
}, llms.WithTools(tools))

choice := resp.Choices[0]
// choice.ReasoningContent is populated correctly
// choice.ToolCalls has the tool call

// Step 2: fails with HTTP 400 because reasoning_content cannot be passed back
assistantMsg := llms.MessageContent{
    Role:  llms.ChatMessageTypeAI,
    Parts: []llms.ContentPart{choice.ToolCalls[0]},
    // ReasoningContent: choice.ReasoningContent  <-- this field doesn't exist on MessageContent!
}

Expected Behavior

MessageContent should support a ReasoningContent field so that reasoning content from API responses can be preserved in conversation history for subsequent requests.

Environment

• langchaingo version: latest main
• Model: deepseek-reasoner
• Related PRs: Add ReasoningContent field for deepseek-reasoner model #1121 (added ReasoningContent parsing), llms: implement StreamingReasoningFunc for reasoning models #1125 (streaming support)
• Same class of bug existed in Python LangChain: [langchain-deepseek] Missing reasoning_content in request payload when using deepseek-reasoner with tool calling langchain-ai/langchain#34166