dsh-ocr-vision: A Local RapidOCR Text Extraction Tool

`dsh-ocr-vision` is a plugin for DeepSeek Harness designed to address the image understanding limitations of pure text models. It adds a local `ocr_image` tool that uses the RapidOCR engine to convert images into plain text, enabling the model to “understand” images. The plugin can be installed directly from GitHub or built locally, and requires Python and the RapidOCR dependencies to be installed on the host machine. Functionally, it supports reading images via `ctx.fs`, region cropping, and parameter configuration. Once installed, the model can directly invoke this tool to perform localized, privacy-preserving OCR processing without uploading images to any API.