Add project-root context binding for agent identification

[sebastientaggart]

Summary

Currently agents are identified by arbitrary IDs (mock-1, mock-2). When running multiple instances of the same agent type (e.g., three Claude Code sessions), there's no way to distinguish which project each is working on.

Proposal

• Add project_root (and optionally active_file) to agent registration metadata
• Agent IDs should incorporate or be mappable to their workspace context
• The OpenDeck plugin should display context-aware labels (e.g., "Claude: feature-x") rather than generic IDs
• The Bridge should track which instance is "active" for a given project

Why

This is the "which copy?" problem — if you have three terminals running the same agent, the dashboard needs to show which one needs attention and for which project. Binding agents to their workspace context makes multi-instance usage practical.

References

Identified during architectural review comparing against the Mediator/Bridge pattern for multi-agent orchestration.

[sebastientaggart]

Resuming work. Implementing project-root context binding: adding metadata fields to AgentBase, updating API endpoints, Bridge client, and OpenDeck plugin display.