dsh-vision-mix: Adding image recognition, image follow-up, and image generation/editing capabilities to DSH.

`dsh-vision-mix` is a visual enhancement plugin for DeepSeek Harness (MIT licensed) designed to provide visual capabilities to Agents that support only text models. By integrating text models, visual models, and optional image generation APIs through the `Mix` model entry point, the plugin enables image recognition, multi-turn follow-up questions, screenshot analysis, and image generation and editing functions. Its core features include detecting and declaring the `image` capability in "Settings → Vision Mix", and supporting `vision_mix_image_generate` (image generation) and `vision_mix_image_edit` (editing based on existing images in the session) via an `OpenAI-compatible` API. Credentials are managed by DS.