Add Configurable Streaming Mode for LLM Providers with Tool-Calling Limitations

[susesamu]

Add a configurable option to disable streaming mode for LLM requests. This would allow Rancher AI Agent to work with providers and proxies that support tool calling only in non-streaming mode.

Problem Statement

The Rancher AI Agent currently sends requests using streaming mode when tool calling is enabled. Some model providers and proxy implementations do not support the combination of streaming and tool use, resulting in request failures.

Example error:

{
  "code": 400,
  "message": "This model doesn't support tool use in streaming mode."
}

In our environment, requests are routed through a GenAI proxy. Due to known issues in the proxy's streaming-mode translator, streaming requests with tools fail, while equivalent non-streaming requests are supported.

At present, there is no user-facing configuration to disable streaming.

Additional Info

JIRA: SURE-11703