feat: redesign World Lab operator workspace

[christopherjnelson]

Summary

• reorganize World Lab into a persistent operator shell with Live and Agents workspaces
• keep one mounted execution controller while adding a compact command bar, contextual inspector tabs, agent rail, and bounded activity dock
• route model, reasoning, personality, and strategy configuration through the dedicated Agents workspace and retain server-authoritative mutations
• update responsive behavior, component coverage, Playwright coverage, and relevant architecture, testing, roadmap, and README documentation

Why

Long-running experiments need a cohesive operator console where run state, agent activity, authoritative territory, recovery telemetry, and configuration remain easy to inspect without interrupting browser-owned playback.

Impact

Operators can switch between live observation and agent configuration without creating a second playback controller or losing active run state. Secondary and destructive actions are consolidated in the overflow menu, while existing world, export, reset, retry, skip, reconciliation, and mutation behavior remains intact.

Validation

• pnpm --filter @agentborne/world-lab test (repository owner confirmed the focused suite after iterative fixes)
• pnpm test:e2e (1 passed)