DSH-Vision-Sidecar: Adding Managed Visual Perception to DeepSeek Harness

`dsh-vision-sidecar` is a DeepSeek Harness (DSH) plugin maintained by `121103qwq`, designed to provide hosted visual perception capabilities for pure-text reasoning models while preserving the user's original inference configuration. This plugin processes images by calling an OpenAI-compatible vision endpoint and returns the generated descriptions as plain text to the model, thereby avoiding the GPU resource requirements and model replacement costs associated with locally deploying vision-language models. By default, the plugin uses LLM7.io's anonymous vision route, which requires no API key or local GPU, but users should be aware of its request rate and token limits. Users can also configure custom OpenAI-compatible vision providers, such as OVHcloud or OpenRo