AI assistant now works with DeepSeek's thinking models

The AI assistant now works with DeepSeek's thinking models, so requests succeed after the URL, API key, and model list all validate.

Studio's OpenAI-compatible adapter (used for DeepSeek) replayed assistant reasoning under OpenRouter field names (reasoning / reasoning_details) for every target. DeepSeek's V4 models run in thinking mode by default and expect that reasoning on reasoning_content, and they reject tool-continuation turns that lack it with HTTP 400 invalid_request_error (for example "Missing reasoning_content field in the assistant message at message index N").