dsh-vision-local: Add local visual capabilities to text-only DeepSeek Harness

dsh-vision-local is a plugin designed specifically for DeepSeek Harness (DSH) to enable pure text models to acquire image understanding capabilities through local vision models, addressing privacy risks and workflow interruptions associated with cloud interfaces. Adopting a "local-first" strategy, the plugin sends images exclusively to local OpenAI-compatible endpoints such as LM Studio and Ollama, without telemetry or additional network calls. Its core advantage lies in returning structured JSON evidence containing a summary, OCR, layout structure, semantics, visual attributes, and uncertain items, effectively suppressing hallucinations. The plugin provides a vision tool and an optional vision-bridge service, supporting the paste or upload of images in formats such as PNG and JPEG, and offering on-demand descriptions. Node.js is required.