Pass context from request endpoint to message handler

从 McpEndpointRouteBuilderExtensions 和 McpServer 开始，跟踪 OnMessageReceivedAsync 以及 issue 中所示的 RequestHandlers.Set 调用。确定如何让不透明的调用方上下文与 IJsonRpcMessage 一起传递到工具处理程序或 MCP 请求上下文中。在长时间运行的工具调用期间可以使用端点提供的上下文，且无需存储或传递已完成的 HttpContext，即表示完成。

Is your feature request related to a problem? Please describe.
When the server is started, it runs in a scope that has the current HttpContext. In that context there might be useful headers that are relevant for that session. But when you need to call a tool the message endpoint might have a different HttpContext and different headers.
For example you might want to use the bearer token from the message endpoint to call a tool or to get the list of tools. In the case of SSE transport, you will need to transfer this token to the tool handler.

Describe the solution you'd like
Be able to build an object (could be opaque) on the message endpoint (can contain anything you want, you can include headers or other info) and pass that object alongside with the IJsonRpcMessage to the tool hander. Maybe make it available on the mcp request context.

Example call:

McpEndpointRouteBuilderExtensions:

await transport.OnMessageReceivedAsync(message, callerContext, context.RequestAborted);

McpServer:

RequestHandlers.Set(
    RequestMethods.ToolsCall,
    (request, callerContext, cancellationToken) => callToolHandler(new(this, request, callerContext), cancellationToken),
    McpJsonUtilities.JsonContext.Default.CallToolRequestParams,
    McpJsonUtilities.JsonContext.Default.CallToolResponse);

This could be extended to other requests as well.

Describe alternatives you've considered
Storing the context in memory introduces additional overhead since it has to be cleaned up and introduces unneeded complexity.
Storing the context in a cache is not ok because we might want to pass sensitive information.
Passing the http context directly is not ok because the post /message request could be finished by the time the tool call is actually processed.

Additional context
In this case we have tool calls with long timeout, they might take a long time to complete, in the order of minutes at least.