How to integrate Docling VLM pipeline with OpenRouter API?

### Question
I would like to integrate Docling’s VlmPipeline with [OpenRouter](https://openrouter.ai/), which provides an OpenAI-compatible API for many hosted models.

My current setup works when targeting a local LM Studio endpoint (though multimodal support there seems limited). Now I want to switch to OpenRouter to leverage its hosted models, but I’m unsure about the correct configuration for ApiVlmOptions.

Start by reading the VlmPipeline and ApiVlmOptions configuration mentioned in the issue, then compare the working local LM Studio setup with OpenRouter’s OpenAI-compatible API. A useful contribution would define and document the correct configuration for hosted multimodal models, including what successful integration looks like.

Written by the indexing model from the issue text.