Feature: Base Multi-agent Collaboration Framework & Shared Context

[Zepan]

Overview

We need to transition PicoClaw from a single-agent architecture to a framework that supports multiple specialized agents. This issue focuses on building the core foundation: defining how agents are structured, how they share information, and how tasks are handed off between them.

Advanced features like Model Routing and Swarm Mode will be tracked in separate follow-up issues.

Core Requirements

• Agent Interface/Struct: Define a standard Agent interface in Go that includes properties like Name, Role, SystemPrompt, and Capabilities.
• Shared Context Pool: Implement a "Shared Memory" or "Blackboard" system where different agents can read from and write to a common session context.
• Basic Hand-off Mechanism: Create a protocol for the "Main Agent" to delegate a specific sub-task to a "Specialized Agent" and receive the output back.
• Agent Lifecycle Management: Basic logic to initialize, activate, and switch between agents within a single user session.

Out of Scope (For this Issue)

• Intelligent Model Routing (Saving tokens via small/large models).
• Swarm Mode (Autonomous agent-to-agent negotiation).
• Visual AIEOS Dashboard.

Why this matters?

This foundation is the prerequisite for all our future "intelligence" features. By getting the base architecture right, we allow the community to build specialized agents (e.g., a "Coder Agent," a "Search Agent," etc.) that can all work together under the PicoClaw umbrella.