ningl32/agent-workbench: Add an observability workbench for DSH's local AI coding agents

ningl32/agent-workbench is a local-first DSH client plugin (plugin name: fde-console / Super Workbench, MIT license) that transforms coding agents such as ZCode, Claude Code, Codex, OpenCode, and DSH into observable and accountable agent teams locally. Core capabilities include a 7-view task panel (Kanban, timeline, acceptance, diary, deep links), real-time observation and transcript readers for six session sources, dispatch prompt restoration, commit attribution based on branch names (task/<id>) or commit messages ([#<id>]) with cross-repository anti-mismatch measures, one-click/batch confirmation of session evidence scored by task, Markdown export for daily and weekly reports, Ctrl+K global search, a 28-day heatmap and SVG scorecard, as well as stdio MCP server, webhooks, Feishu/DingTalk notifications, and Gitea/GitHub/Gitee activity streams. The plugin has zero third-party runtime dependencies; data is stored only in `~/.dsh/fde-console/`. It observes agents in read-only mode, does not persist credentials on disk, redacts output, and routes outbound traffic via an SSRF whitelist. Installation requires Node ≥ 22.13, DSH 0.2.1-rc.2, and `pnpm` in PATH. In addition to installing the plugin itself, you must pin the version to install `@deepseek-ai/dsh-web-app@0.2.1-rc.2`; otherwise, the plugin will load but not activate (pending: waiting for service: webServer). Additionally, it provides `seed-demo.js` for synthetic datasets, `migrate-storage.js` for migration scripts, and `docs/VERIFY.md` for the verification process.