Support PDFs as images on Ollama compatible backends

For providers that do not support the `file` type in the responses API but do allow the `image` type, it should be possible for PDF pages to be automatically converted to images.

The llama.cpp built in server UI already does this, and it could easily be done client-side.

Start by tracing how PDF inputs and the responses API handle providers that reject the `file` type, then compare the llama.cpp built-in server UI behavior. The work is done when PDF pages are automatically converted to `image` inputs for compatible providers that lack `file` support, without changing providers that already accept files.

Written by the indexing model from the issue text.