feat: agent team

[lppp04808]

📝 Description

Implements a multi-agent Teams architecture for PicoClaw, enabling the coordinator agent to decompose complex tasks and delegate to specialized sub-agents running concurrently as goroutines.

Key features added:

• team tool: orchestrate multi-agent pipelines with sequential, parallel, dag, and evaluator_optimizer strategies
• spawn_sub_agent tool: delegate a single isolated task to a sub-agent
• Heterogeneous Agents: per-member model field allows routing tasks to different LLMs (e.g., vision model for screenshots, code model for generation)
• Model tags: annotate models with capability labels (vision, code, fast, long-context, reasoning, image-gen) to guide the coordinator's routing decisions
• Auto-Reviewer: members can declare produces: "code"/"data"/"document" to trigger an automatic QA reviewer after all workers complete
• Soft token budget: budget exhaustion injects a graceful wrap-up signal instead of hard-failing all workers
• Truncation recovery: detects finish_reason=length and injects a retry message instead of looping on malformed JSON

🗣️ Type of Change

• ✨ New feature (non-breaking change which adds functionality)

🤖 AI Code Generation

• 🛠️ Mostly AI-generated (AI draft, Human verified/modified)

🔗 Related Issue

N/A

📚 Technical Context

• Reference URL: N/A
• Reasoning: The Teams architecture enables PicoClaw to handle complex multi-step tasks (e.g., React→Vue migration, full-stack feature development) by decomposing them into specialized sub-agents that run concurrently where possible, with explicit dependency management via DAG strategy.

🧪 Test Environment

• Hardware: PC
• OS: Linux
• Model/Provider:
• Channels: CLI

☑️ Checklist

• My code/docs follow the style of this project.
• I have performed a self-review of my own changes.
• I have updated the documentation accordingly.