Feature Request: Parallel Agent Teams / Multi-Agent Collaboration (like Claude Code Agent Teams)

### What would you like to be added?

Implement a feature in Gemini CLI similar to Claude Code's "Agent Teams" (https://code.claude.com/docs/en/agent-teams), enabling true multi-agent collaboration. The enhancement should support:

- Multiple Gemini agents (local or remote) running in parallel as collaborators, not just as hierarchical sub-agents.
- Agents are able to communicate directly with each other (inter-agent messaging, coordination, sharing findings and challenging results).
- Shared task list/work queue that agents can claim and collaborate on, reflecting more realistic software team workflows.
- Optional team lead/lead agent session to coordinate tasks, but other agents can operate more independently.
- Output displaying the real-time progress/goals of each agent (ideally with session switching in terminal/UI).

This would be a significant expansion beyond the current sub-agent model (which is strictly parent → child delegation with no agent-to-agent messaging or parallel session design).

### Why is this needed?

The current sub-agent model in Gemini CLI is powerful, but remains limited:

- True parallelization is not possible, agents only report back to the main session sequentially.
- Real-world software engineering teams coordinate via direct communication and collaboration, not just top-down delegation.
- The "Agent Teams" model in Claude Code enables faster exploration, competing hypotheses, better debugging, and collaborative coding tasks by allowing agents to work on different components or strategies in parallel and discuss among themselves.
- Adding this would unlock complex, multi-layer project automation and dramatically expand the types of workflows Gemini CLI can support (especially for large teams or automated research/coding agents).

### Additional context

See the Agent Teams concept as described for Claude Code CLI here: https://code.claude.com/docs/en/agent-teams

High-level comparison:
- Claude Code Agent Teams: multi-session, true peer-to-peer agent collaboration, real-time messaging among agents
- Gemini CLI Sub-Agents: hierarchical, no direct agent-to-agent messaging/coordination, no true parallel work

If possible, the Gemini CLI implementation could use the local agent framework plus extension points for shared state, messaging, and session multiplexing.

Happy to help brainstorm use cases and beta test!