Support DeepSeek V4 Flash Vision

Quickly, thank you so much for releasing this and working on this problem and enabling such advanced AI on consumer hardware.

Is there any reason to imagine ds4 cannot handle vision if the underlying DeepSeek model does too?

https://api-docs.deepseek.com/news/news260821/

What would the general steps be to prepare this model for use with ds4?

I'm using a Mac Studio M2 Ultra 192GB - is there any expected gain from omlx, gmlx, or basert given the architecture?

No files, tests, or entry points are named. Start with the linked DeepSeek API documentation and the repository's existing model-input path; completion would require an agreed scope for vision support plus validation on the stated Mac Studio hardware.

索引モデルが issue の本文から書いたものです。