Update Ollama implementation

Currently we treat the Ollama native API as if it was the OpenAI compatible API. We should distinguish and gives users the option between switching between the Ollama native, OpenAI, or Anthropic compatible APIs.

Inspired by this message from Erik Tews: There is one thing that is a bit surprising for me. Rig has a OllamaCompletionRequest with a max_tokens element, but apparently in Ollama there is no such option in the API: https://github.com/ollama/ollama/blob/main/docs/api.md

Look at the current Ollama implementation in the codebase, likely in a module handling API clients. Compare the existing request struct (OllamaCompletionRequest) with the official Ollama API documentation linked in the issue. Identify where API differentiation logic should be added and how to expose the choice between native, OpenAI-compatible, and Anthropic-compatible modes to users. Check for existing tests around API clients to understand the testing pattern.

Vom Indexierungsmodell aus dem Issue-Text verfasst.