dsh-yolo-mode: Add LLM-based automatic approval for DSH sandbox privilege escalation requests

dsh-yolo-mode is an MIT-licensed plugin for DeepSeek Harness (DSH) designed to address the issue where frequent privilege escalation approvals disrupt workflows within sandbox sessions. When the approval strategy is `ask` and the system is in specified sandbox modes (e.g., `workspace-write`), this plugin leverages Large Language Models (LLMs) to automatically adjudicate privilege escalation requests such as `danger-full-access`, thereby replacing manual confirmation steps. It offers presets including `off`, `strict`, `balanced` (default), `permissive`, and `yolo`, and supports configuring custom permission levels via `levels` to set allow, deny, or human-escalation strategies for different target modes or specific tools (e.g., `pwsh`, `write`).