Handle Ollama models that return content in thinking field

[Br1an67]

Summary

Fix Ollama inference failure (Expecting value: line 1 column 1 (char 0)) when models like qwen3-vl return the generated content in the thinking field instead of the response field.

Some Ollama models with thinking mode enabled (e.g., qwen3-vl) place the actual output in the thinking field while leaving response empty. The current code only reads from response, causing json.loads("") to fail.

Closes #992

Changes

• Added a fallback in OllamaService.__call__() to read from the thinking field when response is empty