feat(agents): add council command for multi-model deliberation

[basnijholt]

Summary

• Adds a new council subcommand inspired by Karpathy's llm-council
• Enables collaborative AI deliberation through a 3-stage process:
  1. Stage 1: Query multiple models in parallel
  2. Stage 2: Models peer-review anonymized responses using structured output (JSON mode)
  3. Stage 3: Chairman model synthesizes a final answer

Features

• Works with any OpenAI-compatible API (OpenRouter, Ollama, etc.)
• Structured output for reliable ranking extraction
• Rich console output showing full deliberation process
• --no-ranking flag to skip peer review for faster results
• --final-only flag to show only the synthesis
• Auto-copy final answer to clipboard

Usage

# Basic usage with OpenRouter
agent-cli council "What is the best way to learn programming?" \
    --openai-base-url https://openrouter.ai/api/v1 \
    --openai-api-key sk-or-...

# Custom models
agent-cli council "..." --models "gpt-4o,claude-3-opus,gemini-1.5-pro"

# Skip ranking for faster results
agent-cli council "..." --no-ranking

# Output as JSON
agent-cli council "..." --json

Test plan

• Test with OpenRouter using multiple models
• Test --no-ranking flag skips stage 2
• Test --final-only shows only synthesis
• Test --json output format
• Test clipboard auto-copy